[llvm] 06b9d92 - [X86] Enable v8f16 FABS custom lowering on SSE2 targets

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2023-11-30T10:07:00Z
New Revision: 06b9d92e0e6536d460db3bef6a55de0ecd244927

URL: https://github.com/llvm/llvm-project/commit/06b9d92e0e6536d460db3bef6a55de0ecd244927
DIFF: https://github.com/llvm/llvm-project/commit/06b9d92e0e6536d460db3bef6a55de0ecd244927.diff

LOG: [X86] Enable v8f16 FABS custom lowering on SSE2 targets

Added: 
    

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp
    llvm/test/CodeGen/X86/vec_fabs.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 981f32b77fffad5..587f76f6efb40cc 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1136,6 +1136,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
     setOperationAction(ISD::FMUL, MVT::v8f16, Expand);
     setOperationAction(ISD::FDIV, MVT::v8f16, Expand);
     setOperationAction(ISD::FNEG, MVT::v8f16, Custom);
+    setOperationAction(ISD::FABS, MVT::v8f16, Custom);
 
     // Custom lower v2i64 and v2f64 selects.
     setOperationAction(ISD::SELECT,             MVT::v2f64, Custom);
@@ -1397,8 +1398,6 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
       setOperationAction(ISD::FMINIMUM,          VT, Custom);
     }
 
-    setOperationAction(ISD::FABS, MVT::v8f16, Custom);
-
     // (fp_to_int:v8i16 (v8f32 ..)) requires the result type to be promoted
     // even though v8i16 is a legal type.
     setOperationPromotedToType(ISD::FP_TO_SINT,        MVT::v8i16, MVT::v8i32);

diff  --git a/llvm/test/CodeGen/X86/vec_fabs.ll b/llvm/test/CodeGen/X86/vec_fabs.ll
index c17341c2c8b077e..0fb26ff42d6cec8 100644
--- a/llvm/test/CodeGen/X86/vec_fabs.ll
+++ b/llvm/test/CodeGen/X86/vec_fabs.ll
@@ -141,129 +141,9 @@ declare <4 x float> @llvm.fabs.v4f32(<4 x float> %p)
 define <8 x half> @fabs_v8f16(ptr %p) nounwind {
 ; X86-SSE-LABEL: fabs_v8f16:
 ; X86-SSE:       # %bb.0:
-; X86-SSE-NEXT:    subl $148, %esp
 ; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SSE-NEXT:    movdqa (%eax), %xmm0
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrlq $48, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrld $16, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-SSE-NEXT:    movaps (%eax), %xmm0
 ; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm1, (%esp)
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; X86-SSE-NEXT:    movhps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X86-SSE-NEXT:    # xmm1 = xmm1[0,1],mem[0,1]
-; X86-SSE-NEXT:    movaps %xmm1, %xmm0
-; X86-SSE-NEXT:    addl $148, %esp
 ; X86-SSE-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: fabs_v8f16:
@@ -289,70 +169,8 @@ define <8 x half> @fabs_v8f16(ptr %p) nounwind {
 ;
 ; X64-SSE-LABEL: fabs_v8f16:
 ; X64-SSE:       # %bb.0:
-; X64-SSE-NEXT:    subq $72, %rsp
-; X64-SSE-NEXT:    movdqa (%rdi), %xmm0
-; X64-SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrlq $48, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
+; X64-SSE-NEXT:    movaps (%rdi), %xmm0
 ; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrld $16, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; X64-SSE-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0]
-; X64-SSE-NEXT:    movdqa %xmm1, %xmm0
-; X64-SSE-NEXT:    addq $72, %rsp
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: fabs_v8f16:
@@ -527,251 +345,11 @@ declare <8 x float> @llvm.fabs.v8f32(<8 x float> %p)
 define <16 x half> @fabs_v16f16(ptr %p) nounwind {
 ; X86-SSE-LABEL: fabs_v16f16:
 ; X86-SSE:       # %bb.0:
-; X86-SSE-NEXT:    subl $324, %esp # imm = 0x144
 ; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE-NEXT:    movaps {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN,NaN,NaN,NaN,NaN]
 ; X86-SSE-NEXT:    movaps (%eax), %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqa 16(%eax), %xmm0
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrlq $48, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrld $16, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrlq $48, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrld $16, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm1, (%esp)
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
-; X86-SSE-NEXT:    movdqu %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; X86-SSE-NEXT:    movhps {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X86-SSE-NEXT:    # xmm1 = xmm1[0,1],mem[0,1]
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    addl $324, %esp # imm = 0x144
+; X86-SSE-NEXT:    andps %xmm1, %xmm0
+; X86-SSE-NEXT:    andps 16(%eax), %xmm1
 ; X86-SSE-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: fabs_v16f16:
@@ -797,134 +375,10 @@ define <16 x half> @fabs_v16f16(ptr %p) nounwind {
 ;
 ; X64-SSE-LABEL: fabs_v16f16:
 ; X64-SSE:       # %bb.0:
-; X64-SSE-NEXT:    subq $88, %rsp
-; X64-SSE-NEXT:    movdqa (%rdi), %xmm1
-; X64-SSE-NEXT:    movdqa %xmm1, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movaps 16(%rdi), %xmm0
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa %xmm1, %xmm0
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrlq $48, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrld $16, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; X64-SSE-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0]
-; X64-SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd (%rsp), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    punpckldq (%rsp), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; X64-SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrlq $48, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrld $16, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; X64-SSE-NEXT:    punpcklqdq (%rsp), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0]
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    addq $88, %rsp
+; X64-SSE-NEXT:    movaps {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN,NaN,NaN,NaN,NaN]
+; X64-SSE-NEXT:    movaps (%rdi), %xmm0
+; X64-SSE-NEXT:    andps %xmm1, %xmm0
+; X64-SSE-NEXT:    andps 16(%rdi), %xmm1
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: fabs_v16f16:
@@ -1103,497 +557,15 @@ declare <16 x float> @llvm.fabs.v16f32(<16 x float> %p)
 define <32 x half> @fabs_v32f16(ptr %p) nounwind {
 ; X86-SSE-LABEL: fabs_v32f16:
 ; X86-SSE:       # %bb.0:
-; X86-SSE-NEXT:    subl $644, %esp # imm = 0x284
 ; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE-NEXT:    movaps {{.*#+}} xmm3 = [NaN,NaN,NaN,NaN,NaN,NaN,NaN,NaN]
 ; X86-SSE-NEXT:    movaps (%eax), %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movaps 16(%eax), %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movaps 32(%eax), %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqa 48(%eax), %xmm0
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrlq $48, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrld $16, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrlq $48, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrld $16, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrlq $48, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrld $16, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrlq $48, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    psrld $16, %xmm0
-; X86-SSE-NEXT:    pextrw $0, %xmm0, %eax
-; X86-SSE-NEXT:    movw %ax, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __extendhfsf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm0, (%esp)
-; X86-SSE-NEXT:    fstps {{[0-9]+}}(%esp)
-; X86-SSE-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    movss %xmm1, (%esp)
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0]
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0]
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
-; X86-SSE-NEXT:    movdqu %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; X86-SSE-NEXT:    movdqu %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill
-; X86-SSE-NEXT:    calll __truncsfhf2
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm3 # 16-byte Reload
-; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
-; X86-SSE-NEXT:    movdqu {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; X86-SSE-NEXT:    movhps {{[-0-9]+}}(%e{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; X86-SSE-NEXT:    # xmm3 = xmm3[0,1],mem[0,1]
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm1 # 16-byte Reload
-; X86-SSE-NEXT:    movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 16-byte Reload
-; X86-SSE-NEXT:    addl $644, %esp # imm = 0x284
+; X86-SSE-NEXT:    andps %xmm3, %xmm0
+; X86-SSE-NEXT:    movaps 16(%eax), %xmm1
+; X86-SSE-NEXT:    andps %xmm3, %xmm1
+; X86-SSE-NEXT:    movaps 32(%eax), %xmm2
+; X86-SSE-NEXT:    andps %xmm3, %xmm2
+; X86-SSE-NEXT:    andps 48(%eax), %xmm3
 ; X86-SSE-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: fabs_v32f16:
@@ -1637,262 +609,14 @@ define <32 x half> @fabs_v32f16(ptr %p) nounwind {
 ;
 ; X64-SSE-LABEL: fabs_v32f16:
 ; X64-SSE:       # %bb.0:
-; X64-SSE-NEXT:    subq $120, %rsp
-; X64-SSE-NEXT:    movdqa (%rdi), %xmm1
-; X64-SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps 16(%rdi), %xmm0
-; X64-SSE-NEXT:    movaps %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movaps 32(%rdi), %xmm0
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps 48(%rdi), %xmm0
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa %xmm1, %xmm0
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrlq $48, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrld $16, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; X64-SSE-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0]
-; X64-SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrlq $48, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrld $16, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; X64-SSE-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0]
-; X64-SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd (%rsp), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    punpckldq (%rsp), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrlq $48, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd (%rsp), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movaps %xmm0, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrld $16, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa (%rsp), %xmm1 # 16-byte Reload
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; X64-SSE-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm1 = xmm1[0],mem[0]
-; X64-SSE-NEXT:    movdqa %xmm1, (%rsp) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrlq $48, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    punpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; X64-SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    psrld $16, %xmm0
-; X64-SSE-NEXT:    callq __extendhfsf2 at PLT
-; X64-SSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-SSE-NEXT:    callq __truncsfhf2 at PLT
-; X64-SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
-; X64-SSE-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm3 = xmm3[0],mem[0],xmm3[1],mem[1]
-; X64-SSE-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; X64-SSE-NEXT:    # xmm3 = xmm3[0],mem[0]
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; X64-SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; X64-SSE-NEXT:    movaps (%rsp), %xmm2 # 16-byte Reload
-; X64-SSE-NEXT:    addq $120, %rsp
+; X64-SSE-NEXT:    movaps {{.*#+}} xmm3 = [NaN,NaN,NaN,NaN,NaN,NaN,NaN,NaN]
+; X64-SSE-NEXT:    movaps (%rdi), %xmm0
+; X64-SSE-NEXT:    andps %xmm3, %xmm0
+; X64-SSE-NEXT:    movaps 16(%rdi), %xmm1
+; X64-SSE-NEXT:    andps %xmm3, %xmm1
+; X64-SSE-NEXT:    movaps 32(%rdi), %xmm2
+; X64-SSE-NEXT:    andps %xmm3, %xmm2
+; X64-SSE-NEXT:    andps 48(%rdi), %xmm3
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: fabs_v32f16: