Index: llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
===================================================================
--- llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -918,11 +918,18 @@
   // TODO: Vector types.
   getActionDefinitionsBuilder({G_SADDSAT, G_SSUBSAT}).lowerIf(isScalar(0));
 
-  // TODO: Vector types.
   getActionDefinitionsBuilder({G_FMAXNUM, G_FMINNUM})
-      .legalFor({MinFPScalar, s32, s64})
+      .legalFor({MinFPScalar, s32, s64, v2s32, v4s32, v2s64})
+      .legalIf([=](const LegalityQuery &Query) {
+        const auto &Ty = Query.Types[0];
+        return (Ty == v8s16 || Ty == v4s16) && HasFP16;
+      })
       .libcallFor({s128})
-      .minScalar(0, MinFPScalar);
+      .minScalarOrElt(0, MinFPScalar)
+      .clampNumElements(0, v4s16, v8s16)
+      .clampNumElements(0, v2s32, v4s32)
+      .clampNumElements(0, v2s64, v2s64)
+      .moreElementsToNextPow2(0);
 
   getActionDefinitionsBuilder({G_FMAXIMUM, G_FMINIMUM})
       .legalFor({MinFPScalar, s32, s64, v2s32, v4s32, v2s64})
Index: llvm/test/CodeGen/AArch64/GlobalISel/legalize-fminnum.mir
===================================================================
--- llvm/test/CodeGen/AArch64/GlobalISel/legalize-fminnum.mir
+++ llvm/test/CodeGen/AArch64/GlobalISel/legalize-fminnum.mir
@@ -1,6 +1,6 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=aarch64-unknown-unknown -run-pass=legalizer -mattr=+fullfp16 -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=FP16
-# RUN: llc -mtriple=aarch64-unknown-unknown -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=NO-FP16
+# RUN: llc -mtriple=aarch64-unknown-unknown -run-pass=legalizer -mattr=+fullfp16 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-FP16
+# RUN: llc -mtriple=aarch64-unknown-unknown -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-NOFP16
 ...
 ---
 name:            s16_legal_with_full_fp16
@@ -8,21 +8,26 @@
 body:             |
   bb.0:
     liveins: $h0, $h1
-    ; FP16-LABEL: name: s16_legal_with_full_fp16
-    ; FP16: %a:_(s16) = COPY $h0
-    ; FP16: %b:_(s16) = COPY $h1
-    ; FP16: %minnum:_(s16) = G_FMINNUM %a, %b
-    ; FP16: $h0 = COPY %minnum(s16)
-    ; FP16: RET_ReallyLR implicit $h0
-    ; NO-FP16-LABEL: name: s16_legal_with_full_fp16
-    ; NO-FP16: %a:_(s16) = COPY $h0
-    ; NO-FP16: %b:_(s16) = COPY $h1
-    ; NO-FP16: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT %a(s16)
-    ; NO-FP16: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT %b(s16)
-    ; NO-FP16: [[FMINNUM:%[0-9]+]]:_(s32) = G_FMINNUM [[FPEXT]], [[FPEXT1]]
-    ; NO-FP16: %minnum:_(s16) = G_FPTRUNC [[FMINNUM]](s32)
-    ; NO-FP16: $h0 = COPY %minnum(s16)
-    ; NO-FP16: RET_ReallyLR implicit $h0
+    ; CHECK-FP16-LABEL: name: s16_legal_with_full_fp16
+    ; CHECK-FP16: liveins: $h0, $h1
+    ; CHECK-FP16-NEXT: {{  $}}
+    ; CHECK-FP16-NEXT: %a:_(s16) = COPY $h0
+    ; CHECK-FP16-NEXT: %b:_(s16) = COPY $h1
+    ; CHECK-FP16-NEXT: %minnum:_(s16) = G_FMINNUM %a, %b
+    ; CHECK-FP16-NEXT: $h0 = COPY %minnum(s16)
+    ; CHECK-FP16-NEXT: RET_ReallyLR implicit $h0
+    ;
+    ; CHECK-NOFP16-LABEL: name: s16_legal_with_full_fp16
+    ; CHECK-NOFP16: liveins: $h0, $h1
+    ; CHECK-NOFP16-NEXT: {{  $}}
+    ; CHECK-NOFP16-NEXT: %a:_(s16) = COPY $h0
+    ; CHECK-NOFP16-NEXT: %b:_(s16) = COPY $h1
+    ; CHECK-NOFP16-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT %a(s16)
+    ; CHECK-NOFP16-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT %b(s16)
+    ; CHECK-NOFP16-NEXT: [[FMINNUM:%[0-9]+]]:_(s32) = G_FMINNUM [[FPEXT]], [[FPEXT1]]
+    ; CHECK-NOFP16-NEXT: %minnum:_(s16) = G_FPTRUNC [[FMINNUM]](s32)
+    ; CHECK-NOFP16-NEXT: $h0 = COPY %minnum(s16)
+    ; CHECK-NOFP16-NEXT: RET_ReallyLR implicit $h0
     %a:_(s16) = COPY $h0
     %b:_(s16) = COPY $h1
     %minnum:_(s16) = G_FMINNUM %a, %b
@@ -36,18 +41,14 @@
 body:             |
   bb.0:
     liveins: $s0, $s1
-    ; FP16-LABEL: name: s32_legal
-    ; FP16: %a:_(s32) = COPY $s0
-    ; FP16: %b:_(s32) = COPY $s1
-    ; FP16: %minnum:_(s32) = G_FMINNUM %a, %b
-    ; FP16: $s0 = COPY %minnum(s32)
-    ; FP16: RET_ReallyLR implicit $s0
-    ; NO-FP16-LABEL: name: s32_legal
-    ; NO-FP16: %a:_(s32) = COPY $s0
-    ; NO-FP16: %b:_(s32) = COPY $s1
-    ; NO-FP16: %minnum:_(s32) = G_FMINNUM %a, %b
-    ; NO-FP16: $s0 = COPY %minnum(s32)
-    ; NO-FP16: RET_ReallyLR implicit $s0
+    ; CHECK-LABEL: name: s32_legal
+    ; CHECK: liveins: $s0, $s1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %a:_(s32) = COPY $s0
+    ; CHECK-NEXT: %b:_(s32) = COPY $s1
+    ; CHECK-NEXT: %minnum:_(s32) = G_FMINNUM %a, %b
+    ; CHECK-NEXT: $s0 = COPY %minnum(s32)
+    ; CHECK-NEXT: RET_ReallyLR implicit $s0
     %a:_(s32) = COPY $s0
     %b:_(s32) = COPY $s1
     %minnum:_(s32) = G_FMINNUM %a, %b
@@ -61,18 +62,14 @@
 body:             |
   bb.0:
     liveins: $d0, $d1
-    ; FP16-LABEL: name: s64_legal
-    ; FP16: %a:_(s64) = COPY $d0
-    ; FP16: %b:_(s64) = COPY $d1
-    ; FP16: %minnum:_(s64) = G_FMINNUM %a, %b
-    ; FP16: $d0 = COPY %minnum(s64)
-    ; FP16: RET_ReallyLR implicit $d0
-    ; NO-FP16-LABEL: name: s64_legal
-    ; NO-FP16: %a:_(s64) = COPY $d0
-    ; NO-FP16: %b:_(s64) = COPY $d1
-    ; NO-FP16: %minnum:_(s64) = G_FMINNUM %a, %b
-    ; NO-FP16: $d0 = COPY %minnum(s64)
-    ; NO-FP16: RET_ReallyLR implicit $d0
+    ; CHECK-LABEL: name: s64_legal
+    ; CHECK: liveins: $d0, $d1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %a:_(s64) = COPY $d0
+    ; CHECK-NEXT: %b:_(s64) = COPY $d1
+    ; CHECK-NEXT: %minnum:_(s64) = G_FMINNUM %a, %b
+    ; CHECK-NEXT: $d0 = COPY %minnum(s64)
+    ; CHECK-NEXT: RET_ReallyLR implicit $d0
     %a:_(s64) = COPY $d0
     %b:_(s64) = COPY $d1
     %minnum:_(s64) = G_FMINNUM %a, %b
@@ -86,30 +83,66 @@
 body:             |
   bb.0:
     liveins: $q0, $q1
-    ; FP16-LABEL: name: s128_libcall
-    ; FP16: %a:_(s128) = COPY $q0
-    ; FP16: %b:_(s128) = COPY $q1
-    ; FP16: ADJCALLSTACKDOWN 0, 0, implicit-def $sp, implicit $sp
-    ; FP16: $q0 = COPY %a(s128)
-    ; FP16: $q1 = COPY %b(s128)
-    ; FP16: BL &fminl, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $q0, implicit $q1, implicit-def $q0
-    ; FP16: ADJCALLSTACKUP 0, 0, implicit-def $sp, implicit $sp
-    ; FP16: %minnum:_(s128) = COPY $q0
-    ; FP16: $q0 = COPY %minnum(s128)
-    ; FP16: RET_ReallyLR implicit $q0
-    ; NO-FP16-LABEL: name: s128_libcall
-    ; NO-FP16: %a:_(s128) = COPY $q0
-    ; NO-FP16: %b:_(s128) = COPY $q1
-    ; NO-FP16: ADJCALLSTACKDOWN 0, 0, implicit-def $sp, implicit $sp
-    ; NO-FP16: $q0 = COPY %a(s128)
-    ; NO-FP16: $q1 = COPY %b(s128)
-    ; NO-FP16: BL &fminl, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $q0, implicit $q1, implicit-def $q0
-    ; NO-FP16: ADJCALLSTACKUP 0, 0, implicit-def $sp, implicit $sp
-    ; NO-FP16: %minnum:_(s128) = COPY $q0
-    ; NO-FP16: $q0 = COPY %minnum(s128)
-    ; NO-FP16: RET_ReallyLR implicit $q0
+    ; CHECK-LABEL: name: s128_libcall
+    ; CHECK: liveins: $q0, $q1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %a:_(s128) = COPY $q0
+    ; CHECK-NEXT: %b:_(s128) = COPY $q1
+    ; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $sp, implicit $sp
+    ; CHECK-NEXT: $q0 = COPY %a(s128)
+    ; CHECK-NEXT: $q1 = COPY %b(s128)
+    ; CHECK-NEXT: BL &fminl, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $q0, implicit $q1, implicit-def $q0
+    ; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def $sp, implicit $sp
+    ; CHECK-NEXT: %minnum:_(s128) = COPY $q0
+    ; CHECK-NEXT: $q0 = COPY %minnum(s128)
+    ; CHECK-NEXT: RET_ReallyLR implicit $q0
     %a:_(s128) = COPY $q0
     %b:_(s128) = COPY $q1
     %minnum:_(s128) = G_FMINNUM %a, %b
     $q0 = COPY %minnum(s128)
     RET_ReallyLR implicit $q0
+
+...
+---
+name:            v4s32_legal
+alignment:       4
+body:             |
+  bb.0:
+    liveins: $q0, $q1
+    ; CHECK-LABEL: name: v4s32_legal
+    ; CHECK: liveins: $q0, $q1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %a:_(<4 x s32>) = COPY $q0
+    ; CHECK-NEXT: %b:_(<4 x s32>) = COPY $q1
+    ; CHECK-NEXT: %minnum:_(<4 x s32>) = G_FMINNUM %a, %b
+    ; CHECK-NEXT: $q0 = COPY %minnum(<4 x s32>)
+    ; CHECK-NEXT: RET_ReallyLR implicit $q0
+    %a:_(<4 x s32>) = COPY $q0
+    %b:_(<4 x s32>) = COPY $q1
+    %minnum:_(<4 x s32>) = G_FMINNUM %a, %b
+    $q0 = COPY %minnum(<4 x s32>)
+    RET_ReallyLR implicit $q0
+
+...
+---
+name:            v3s32_widen
+alignment:       4
+body:             |
+  bb.1.entry:
+    liveins: $q0, $q1
+    %2:_(<2 x s64>) = COPY $q0
+    %3:_(<4 x s32>) = G_BITCAST %2:_(<2 x s64>)
+    %4:_(s32), %5:_(s32), %6:_(s32), %7:_(s32) = G_UNMERGE_VALUES %3:_(<4 x s32>)
+    %0:_(<3 x s32>) = G_BUILD_VECTOR %4:_(s32), %5:_(s32), %6:_(s32)
+    %8:_(<2 x s64>) = COPY $q1
+    %9:_(<4 x s32>) = G_BITCAST %8:_(<2 x s64>)
+    %10:_(s32), %11:_(s32), %12:_(s32), %13:_(s32) = G_UNMERGE_VALUES %9:_(<4 x s32>)
+    %1:_(<3 x s32>) = G_BUILD_VECTOR %10:_(s32), %11:_(s32), %12:_(s32)
+    %14:_(<3 x s32>) = G_FMINNUM %0:_, %1:_
+    %15:_(s32), %16:_(s32), %17:_(s32) = G_UNMERGE_VALUES %14:_(<3 x s32>)
+    %18:_(s32) = G_IMPLICIT_DEF
+    %19:_(<4 x s32>) = G_BUILD_VECTOR %15:_(s32), %16:_(s32), %17:_(s32), %18:_(s32)
+    $q0 = COPY %19:_(<4 x s32>)
+    RET_ReallyLR implicit $q0
+
+...
Index: llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
===================================================================
--- llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
+++ llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
@@ -521,11 +521,11 @@
 # DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT: G_FMINNUM (opcode {{[0-9]+}}): 1 type index
 # DEBUG-NEXT: .. opcode {{[0-9]+}} is aliased to {{[0-9]+}}
-# DEBUG-NEXT: .. the first uncovered type index: 1, OK
-# DEBUG-NEXT: .. the first uncovered imm index: 0, OK
+# DEBUG-NEXT: .. type index coverage check SKIPPED: user-defined predicate detected
+# DEBUG-NEXT: .. imm index coverage check SKIPPED: user-defined predicate detected
 # DEBUG-NEXT: G_FMAXNUM (opcode {{[0-9]+}}): 1 type index
-# DEBUG-NEXT: .. the first uncovered type index: 1, OK
-# DEBUG-NEXT: .. the first uncovered imm index: 0, OK
+# DEBUG-NEXT: .. type index coverage check SKIPPED: user-defined predicate detected
+# DEBUG-NEXT: .. imm index coverage check SKIPPED: user-defined predicate detected
 # DEBUG-NEXT: G_FMINNUM_IEEE (opcode {{[0-9]+}}): 1 type index
 # DEBUG-NEXT: .. type index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined
Index: llvm/test/CodeGen/AArch64/fminmax.ll
===================================================================
--- llvm/test/CodeGen/AArch64/fminmax.ll
+++ llvm/test/CodeGen/AArch64/fminmax.ll
@@ -1,6 +1,8 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
-; RUN: llc -mtriple=aarch64-none-eabi -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-NOFP16
-; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-FP16
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
+; RUN: llc -mtriple=aarch64-none-eabi -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-NOFP16-SD
+; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-FP16-SD
+; RUN: llc -mtriple=aarch64-none-eabi -global-isel -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-NOFP16-GI
+; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -global-isel -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-FP16-GI
 
 define double @min_f64(double %a, double %b) {
 ; CHECK-LABEL: min_f64:
@@ -43,36 +45,62 @@
 }
 
 define half @min_f16(half %a, half %b) {
-; CHECK-NOFP16-LABEL: min_f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-NOFP16-NEXT:    fminnm s0, s0, s1
-; CHECK-NOFP16-NEXT:    fcvt h0, s0
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: min_f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fminnm h0, h0, h1
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: min_f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT:    fminnm s0, s0, s1
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: min_f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fminnm h0, h0, h1
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: min_f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-GI-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-GI-NEXT:    fminnm s0, s0, s1
+; CHECK-NOFP16-GI-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: min_f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    fminnm h0, h0, h1
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call half @llvm.minnum.f16(half %a, half %b)
   ret half %c
 }
 
 define half @max_f16(half %a, half %b) {
-; CHECK-NOFP16-LABEL: max_f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-NOFP16-NEXT:    fmaxnm s0, s0, s1
-; CHECK-NOFP16-NEXT:    fcvt h0, s0
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: max_f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fmaxnm h0, h0, h1
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: max_f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s0, s0, s1
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: max_f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fmaxnm h0, h0, h1
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: max_f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-GI-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-GI-NEXT:    fmaxnm s0, s0, s1
+; CHECK-NOFP16-GI-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: max_f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    fmaxnm h0, h0, h1
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call half @llvm.maxnum.f16(half %a, half %b)
   ret half %c
@@ -99,46 +127,74 @@
 }
 
 define <3 x double> @min_v3f64(<3 x double> %a, <3 x double> %b) {
-; CHECK-LABEL: min_v3f64:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    // kill: def $d3 killed $d3 def $q3
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    // kill: def $d4 killed $d4 def $q4
-; CHECK-NEXT:    // kill: def $d1 killed $d1 def $q1
-; CHECK-NEXT:    // kill: def $d2 killed $d2 def $q2
-; CHECK-NEXT:    // kill: def $d5 killed $d5 def $q5
-; CHECK-NEXT:    mov v3.d[1], v4.d[0]
-; CHECK-NEXT:    mov v0.d[1], v1.d[0]
-; CHECK-NEXT:    fminnm v2.2d, v2.2d, v5.2d
-; CHECK-NEXT:    // kill: def $d2 killed $d2 killed $q2
-; CHECK-NEXT:    fminnm v0.2d, v0.2d, v3.2d
-; CHECK-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    // kill: def $d1 killed $d1 killed $q1
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: min_v3f64:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    // kill: def $d3 killed $d3 def $q3
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-SD-NEXT:    // kill: def $d4 killed $d4 def $q4
+; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
+; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 def $q2
+; CHECK-SD-NEXT:    // kill: def $d5 killed $d5 def $q5
+; CHECK-SD-NEXT:    mov v3.d[1], v4.d[0]
+; CHECK-SD-NEXT:    mov v0.d[1], v1.d[0]
+; CHECK-SD-NEXT:    fminnm v2.2d, v2.2d, v5.2d
+; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
+; CHECK-SD-NEXT:    fminnm v0.2d, v0.2d, v3.2d
+; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: min_v3f64:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-GI-NEXT:    // kill: def $d3 killed $d3 def $q3
+; CHECK-GI-NEXT:    // kill: def $d1 killed $d1 def $q1
+; CHECK-GI-NEXT:    // kill: def $d4 killed $d4 def $q4
+; CHECK-GI-NEXT:    fminnm d2, d2, d5
+; CHECK-GI-NEXT:    mov v0.d[1], v1.d[0]
+; CHECK-GI-NEXT:    mov v3.d[1], v4.d[0]
+; CHECK-GI-NEXT:    fminnm v0.2d, v0.2d, v3.2d
+; CHECK-GI-NEXT:    mov d1, v0.d[1]
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT:    ret
 entry:
   %c = call <3 x double> @llvm.minnum.v3f64(<3 x double> %a, <3 x double> %b)
   ret <3 x double> %c
 }
 
 define <3 x double> @max_v3f64(<3 x double> %a, <3 x double> %b) {
-; CHECK-LABEL: max_v3f64:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    // kill: def $d3 killed $d3 def $q3
-; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NEXT:    // kill: def $d4 killed $d4 def $q4
-; CHECK-NEXT:    // kill: def $d1 killed $d1 def $q1
-; CHECK-NEXT:    // kill: def $d2 killed $d2 def $q2
-; CHECK-NEXT:    // kill: def $d5 killed $d5 def $q5
-; CHECK-NEXT:    mov v3.d[1], v4.d[0]
-; CHECK-NEXT:    mov v0.d[1], v1.d[0]
-; CHECK-NEXT:    fmaxnm v2.2d, v2.2d, v5.2d
-; CHECK-NEXT:    // kill: def $d2 killed $d2 killed $q2
-; CHECK-NEXT:    fmaxnm v0.2d, v0.2d, v3.2d
-; CHECK-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    // kill: def $d1 killed $d1 killed $q1
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: max_v3f64:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    // kill: def $d3 killed $d3 def $q3
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-SD-NEXT:    // kill: def $d4 killed $d4 def $q4
+; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
+; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 def $q2
+; CHECK-SD-NEXT:    // kill: def $d5 killed $d5 def $q5
+; CHECK-SD-NEXT:    mov v3.d[1], v4.d[0]
+; CHECK-SD-NEXT:    mov v0.d[1], v1.d[0]
+; CHECK-SD-NEXT:    fmaxnm v2.2d, v2.2d, v5.2d
+; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
+; CHECK-SD-NEXT:    fmaxnm v0.2d, v0.2d, v3.2d
+; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: max_v3f64:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-GI-NEXT:    // kill: def $d3 killed $d3 def $q3
+; CHECK-GI-NEXT:    // kill: def $d1 killed $d1 def $q1
+; CHECK-GI-NEXT:    // kill: def $d4 killed $d4 def $q4
+; CHECK-GI-NEXT:    fmaxnm d2, d2, d5
+; CHECK-GI-NEXT:    mov v0.d[1], v1.d[0]
+; CHECK-GI-NEXT:    mov v3.d[1], v4.d[0]
+; CHECK-GI-NEXT:    fmaxnm v0.2d, v0.2d, v3.2d
+; CHECK-GI-NEXT:    mov d1, v0.d[1]
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT:    ret
 entry:
   %c = call <3 x double> @llvm.maxnum.v3f64(<3 x double> %a, <3 x double> %b)
   ret <3 x double> %c
@@ -187,20 +243,60 @@
 }
 
 define <3 x float> @min_v3f32(<3 x float> %a, <3 x float> %b) {
-; CHECK-LABEL: min_v3f32:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fminnm v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: min_v3f32:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    fminnm v0.4s, v0.4s, v1.4s
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: min_v3f32:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov s2, v0.s[1]
+; CHECK-GI-NEXT:    mov s3, v1.s[1]
+; CHECK-GI-NEXT:    mov s4, v0.s[2]
+; CHECK-GI-NEXT:    mov s5, v1.s[2]
+; CHECK-GI-NEXT:    mov v0.s[1], v2.s[0]
+; CHECK-GI-NEXT:    mov v1.s[1], v3.s[0]
+; CHECK-GI-NEXT:    mov v0.s[2], v4.s[0]
+; CHECK-GI-NEXT:    mov v1.s[2], v5.s[0]
+; CHECK-GI-NEXT:    mov v0.s[3], v0.s[0]
+; CHECK-GI-NEXT:    mov v1.s[3], v0.s[0]
+; CHECK-GI-NEXT:    fminnm v0.4s, v0.4s, v1.4s
+; CHECK-GI-NEXT:    mov s1, v0.s[1]
+; CHECK-GI-NEXT:    mov s2, v0.s[2]
+; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-GI-NEXT:    mov v0.s[3], v0.s[0]
+; CHECK-GI-NEXT:    ret
 entry:
   %c = call <3 x float> @llvm.minnum.v3f32(<3 x float> %a, <3 x float> %b)
   ret <3 x float> %c
 }
 
 define <3 x float> @max_v3f32(<3 x float> %a, <3 x float> %b) {
-; CHECK-LABEL: max_v3f32:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fmaxnm v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: max_v3f32:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    fmaxnm v0.4s, v0.4s, v1.4s
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: max_v3f32:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov s2, v0.s[1]
+; CHECK-GI-NEXT:    mov s3, v1.s[1]
+; CHECK-GI-NEXT:    mov s4, v0.s[2]
+; CHECK-GI-NEXT:    mov s5, v1.s[2]
+; CHECK-GI-NEXT:    mov v0.s[1], v2.s[0]
+; CHECK-GI-NEXT:    mov v1.s[1], v3.s[0]
+; CHECK-GI-NEXT:    mov v0.s[2], v4.s[0]
+; CHECK-GI-NEXT:    mov v1.s[2], v5.s[0]
+; CHECK-GI-NEXT:    mov v0.s[3], v0.s[0]
+; CHECK-GI-NEXT:    mov v1.s[3], v0.s[0]
+; CHECK-GI-NEXT:    fmaxnm v0.4s, v0.4s, v1.4s
+; CHECK-GI-NEXT:    mov s1, v0.s[1]
+; CHECK-GI-NEXT:    mov s2, v0.s[2]
+; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-GI-NEXT:    mov v0.s[3], v0.s[0]
+; CHECK-GI-NEXT:    ret
 entry:
   %c = call <3 x float> @llvm.maxnum.v3f32(<3 x float> %a, <3 x float> %b)
   ret <3 x float> %c
@@ -227,84 +323,162 @@
 }
 
 define <7 x float> @min_v7f32(<7 x float> %a, <7 x float> %b) {
-; CHECK-LABEL: min_v7f32:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    mov x8, sp
-; CHECK-NEXT:    // kill: def $s7 killed $s7 def $q7
-; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
-; CHECK-NEXT:    // kill: def $s1 killed $s1 def $q1
-; CHECK-NEXT:    add x9, sp, #32
-; CHECK-NEXT:    mov v0.s[1], v1.s[0]
-; CHECK-NEXT:    ldr s1, [sp, #24]
-; CHECK-NEXT:    // kill: def $s4 killed $s4 def $q4
-; CHECK-NEXT:    // kill: def $s5 killed $s5 def $q5
-; CHECK-NEXT:    // kill: def $s2 killed $s2 def $q2
-; CHECK-NEXT:    // kill: def $s6 killed $s6 def $q6
-; CHECK-NEXT:    // kill: def $s3 killed $s3 def $q3
-; CHECK-NEXT:    ld1 { v7.s }[1], [x8]
-; CHECK-NEXT:    add x8, sp, #8
-; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
-; CHECK-NEXT:    mov v0.s[2], v2.s[0]
-; CHECK-NEXT:    mov v4.s[1], v5.s[0]
-; CHECK-NEXT:    ld1 { v7.s }[2], [x8]
-; CHECK-NEXT:    add x8, sp, #16
-; CHECK-NEXT:    mov v0.s[3], v3.s[0]
-; CHECK-NEXT:    mov v4.s[2], v6.s[0]
-; CHECK-NEXT:    ld1 { v7.s }[3], [x8]
-; CHECK-NEXT:    add x8, sp, #40
-; CHECK-NEXT:    ld1 { v1.s }[2], [x8]
-; CHECK-NEXT:    fminnm v0.4s, v0.4s, v7.4s
-; CHECK-NEXT:    fminnm v4.4s, v4.4s, v1.4s
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    mov s2, v0.s[2]
-; CHECK-NEXT:    mov s3, v0.s[3]
-; CHECK-NEXT:    // kill: def $s0 killed $s0 killed $q0
-; CHECK-NEXT:    mov s5, v4.s[1]
-; CHECK-NEXT:    mov s6, v4.s[2]
-; CHECK-NEXT:    // kill: def $s4 killed $s4 killed $q4
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: min_v7f32:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    mov x8, sp
+; CHECK-SD-NEXT:    // kill: def $s7 killed $s7 def $q7
+; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-SD-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-SD-NEXT:    add x9, sp, #32
+; CHECK-SD-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-SD-NEXT:    ldr s1, [sp, #24]
+; CHECK-SD-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-SD-NEXT:    // kill: def $s5 killed $s5 def $q5
+; CHECK-SD-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-SD-NEXT:    // kill: def $s6 killed $s6 def $q6
+; CHECK-SD-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-SD-NEXT:    ld1 { v7.s }[1], [x8]
+; CHECK-SD-NEXT:    add x8, sp, #8
+; CHECK-SD-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-SD-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-SD-NEXT:    mov v4.s[1], v5.s[0]
+; CHECK-SD-NEXT:    ld1 { v7.s }[2], [x8]
+; CHECK-SD-NEXT:    add x8, sp, #16
+; CHECK-SD-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-SD-NEXT:    mov v4.s[2], v6.s[0]
+; CHECK-SD-NEXT:    ld1 { v7.s }[3], [x8]
+; CHECK-SD-NEXT:    add x8, sp, #40
+; CHECK-SD-NEXT:    ld1 { v1.s }[2], [x8]
+; CHECK-SD-NEXT:    fminnm v0.4s, v0.4s, v7.4s
+; CHECK-SD-NEXT:    fminnm v4.4s, v4.4s, v1.4s
+; CHECK-SD-NEXT:    mov s1, v0.s[1]
+; CHECK-SD-NEXT:    mov s2, v0.s[2]
+; CHECK-SD-NEXT:    mov s3, v0.s[3]
+; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 killed $q0
+; CHECK-SD-NEXT:    mov s5, v4.s[1]
+; CHECK-SD-NEXT:    mov s6, v4.s[2]
+; CHECK-SD-NEXT:    // kill: def $s4 killed $s4 killed $q4
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: min_v7f32:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-GI-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-GI-NEXT:    ldr s16, [sp]
+; CHECK-GI-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-GI-NEXT:    // kill: def $s7 killed $s7 def $q7
+; CHECK-GI-NEXT:    // kill: def $s5 killed $s5 def $q5
+; CHECK-GI-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-GI-NEXT:    // kill: def $s6 killed $s6 def $q6
+; CHECK-GI-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT:    ldr s19, [sp, #24]
+; CHECK-GI-NEXT:    ldr s20, [sp, #32]
+; CHECK-GI-NEXT:    mov v7.s[1], v16.s[0]
+; CHECK-GI-NEXT:    ldr s17, [sp, #8]
+; CHECK-GI-NEXT:    mov v4.s[1], v5.s[0]
+; CHECK-GI-NEXT:    ldr s21, [sp, #40]
+; CHECK-GI-NEXT:    mov v19.s[1], v20.s[0]
+; CHECK-GI-NEXT:    ldr s18, [sp, #16]
+; CHECK-GI-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-GI-NEXT:    mov v7.s[2], v17.s[0]
+; CHECK-GI-NEXT:    mov v4.s[2], v6.s[0]
+; CHECK-GI-NEXT:    mov v19.s[2], v21.s[0]
+; CHECK-GI-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-GI-NEXT:    mov v7.s[3], v18.s[0]
+; CHECK-GI-NEXT:    mov v4.s[3], v0.s[0]
+; CHECK-GI-NEXT:    mov v19.s[3], v0.s[0]
+; CHECK-GI-NEXT:    fminnm v0.4s, v0.4s, v7.4s
+; CHECK-GI-NEXT:    fminnm v4.4s, v4.4s, v19.4s
+; CHECK-GI-NEXT:    mov s1, v0.s[1]
+; CHECK-GI-NEXT:    mov s2, v0.s[2]
+; CHECK-GI-NEXT:    mov s3, v0.s[3]
+; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 killed $q0
+; CHECK-GI-NEXT:    mov s5, v4.s[1]
+; CHECK-GI-NEXT:    mov s6, v4.s[2]
+; CHECK-GI-NEXT:    // kill: def $s4 killed $s4 killed $q4
+; CHECK-GI-NEXT:    ret
 entry:
   %c = call <7 x float> @llvm.minnum.v7f32(<7 x float> %a, <7 x float> %b)
   ret <7 x float> %c
 }
 
 define <7 x float> @max_v7f32(<7 x float> %a, <7 x float> %b) {
-; CHECK-LABEL: max_v7f32:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    mov x8, sp
-; CHECK-NEXT:    // kill: def $s7 killed $s7 def $q7
-; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
-; CHECK-NEXT:    // kill: def $s1 killed $s1 def $q1
-; CHECK-NEXT:    add x9, sp, #32
-; CHECK-NEXT:    mov v0.s[1], v1.s[0]
-; CHECK-NEXT:    ldr s1, [sp, #24]
-; CHECK-NEXT:    // kill: def $s4 killed $s4 def $q4
-; CHECK-NEXT:    // kill: def $s5 killed $s5 def $q5
-; CHECK-NEXT:    // kill: def $s2 killed $s2 def $q2
-; CHECK-NEXT:    // kill: def $s6 killed $s6 def $q6
-; CHECK-NEXT:    // kill: def $s3 killed $s3 def $q3
-; CHECK-NEXT:    ld1 { v7.s }[1], [x8]
-; CHECK-NEXT:    add x8, sp, #8
-; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
-; CHECK-NEXT:    mov v0.s[2], v2.s[0]
-; CHECK-NEXT:    mov v4.s[1], v5.s[0]
-; CHECK-NEXT:    ld1 { v7.s }[2], [x8]
-; CHECK-NEXT:    add x8, sp, #16
-; CHECK-NEXT:    mov v0.s[3], v3.s[0]
-; CHECK-NEXT:    mov v4.s[2], v6.s[0]
-; CHECK-NEXT:    ld1 { v7.s }[3], [x8]
-; CHECK-NEXT:    add x8, sp, #40
-; CHECK-NEXT:    ld1 { v1.s }[2], [x8]
-; CHECK-NEXT:    fmaxnm v0.4s, v0.4s, v7.4s
-; CHECK-NEXT:    fmaxnm v4.4s, v4.4s, v1.4s
-; CHECK-NEXT:    mov s1, v0.s[1]
-; CHECK-NEXT:    mov s2, v0.s[2]
-; CHECK-NEXT:    mov s3, v0.s[3]
-; CHECK-NEXT:    // kill: def $s0 killed $s0 killed $q0
-; CHECK-NEXT:    mov s5, v4.s[1]
-; CHECK-NEXT:    mov s6, v4.s[2]
-; CHECK-NEXT:    // kill: def $s4 killed $s4 killed $q4
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: max_v7f32:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    mov x8, sp
+; CHECK-SD-NEXT:    // kill: def $s7 killed $s7 def $q7
+; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-SD-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-SD-NEXT:    add x9, sp, #32
+; CHECK-SD-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-SD-NEXT:    ldr s1, [sp, #24]
+; CHECK-SD-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-SD-NEXT:    // kill: def $s5 killed $s5 def $q5
+; CHECK-SD-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-SD-NEXT:    // kill: def $s6 killed $s6 def $q6
+; CHECK-SD-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-SD-NEXT:    ld1 { v7.s }[1], [x8]
+; CHECK-SD-NEXT:    add x8, sp, #8
+; CHECK-SD-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-SD-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-SD-NEXT:    mov v4.s[1], v5.s[0]
+; CHECK-SD-NEXT:    ld1 { v7.s }[2], [x8]
+; CHECK-SD-NEXT:    add x8, sp, #16
+; CHECK-SD-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-SD-NEXT:    mov v4.s[2], v6.s[0]
+; CHECK-SD-NEXT:    ld1 { v7.s }[3], [x8]
+; CHECK-SD-NEXT:    add x8, sp, #40
+; CHECK-SD-NEXT:    ld1 { v1.s }[2], [x8]
+; CHECK-SD-NEXT:    fmaxnm v0.4s, v0.4s, v7.4s
+; CHECK-SD-NEXT:    fmaxnm v4.4s, v4.4s, v1.4s
+; CHECK-SD-NEXT:    mov s1, v0.s[1]
+; CHECK-SD-NEXT:    mov s2, v0.s[2]
+; CHECK-SD-NEXT:    mov s3, v0.s[3]
+; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 killed $q0
+; CHECK-SD-NEXT:    mov s5, v4.s[1]
+; CHECK-SD-NEXT:    mov s6, v4.s[2]
+; CHECK-SD-NEXT:    // kill: def $s4 killed $s4 killed $q4
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: max_v7f32:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
+; CHECK-GI-NEXT:    // kill: def $s1 killed $s1 def $q1
+; CHECK-GI-NEXT:    ldr s16, [sp]
+; CHECK-GI-NEXT:    // kill: def $s4 killed $s4 def $q4
+; CHECK-GI-NEXT:    // kill: def $s7 killed $s7 def $q7
+; CHECK-GI-NEXT:    // kill: def $s5 killed $s5 def $q5
+; CHECK-GI-NEXT:    // kill: def $s2 killed $s2 def $q2
+; CHECK-GI-NEXT:    // kill: def $s6 killed $s6 def $q6
+; CHECK-GI-NEXT:    // kill: def $s3 killed $s3 def $q3
+; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT:    ldr s19, [sp, #24]
+; CHECK-GI-NEXT:    ldr s20, [sp, #32]
+; CHECK-GI-NEXT:    mov v7.s[1], v16.s[0]
+; CHECK-GI-NEXT:    ldr s17, [sp, #8]
+; CHECK-GI-NEXT:    mov v4.s[1], v5.s[0]
+; CHECK-GI-NEXT:    ldr s21, [sp, #40]
+; CHECK-GI-NEXT:    mov v19.s[1], v20.s[0]
+; CHECK-GI-NEXT:    ldr s18, [sp, #16]
+; CHECK-GI-NEXT:    mov v0.s[2], v2.s[0]
+; CHECK-GI-NEXT:    mov v7.s[2], v17.s[0]
+; CHECK-GI-NEXT:    mov v4.s[2], v6.s[0]
+; CHECK-GI-NEXT:    mov v19.s[2], v21.s[0]
+; CHECK-GI-NEXT:    mov v0.s[3], v3.s[0]
+; CHECK-GI-NEXT:    mov v7.s[3], v18.s[0]
+; CHECK-GI-NEXT:    mov v4.s[3], v0.s[0]
+; CHECK-GI-NEXT:    mov v19.s[3], v0.s[0]
+; CHECK-GI-NEXT:    fmaxnm v0.4s, v0.4s, v7.4s
+; CHECK-GI-NEXT:    fmaxnm v4.4s, v4.4s, v19.4s
+; CHECK-GI-NEXT:    mov s1, v0.s[1]
+; CHECK-GI-NEXT:    mov s2, v0.s[2]
+; CHECK-GI-NEXT:    mov s3, v0.s[3]
+; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 killed $q0
+; CHECK-GI-NEXT:    mov s5, v4.s[1]
+; CHECK-GI-NEXT:    mov s6, v4.s[2]
+; CHECK-GI-NEXT:    // kill: def $s4 killed $s4 killed $q4
+; CHECK-GI-NEXT:    ret
 entry:
   %c = call <7 x float> @llvm.maxnum.v7f32(<7 x float> %a, <7 x float> %b)
   ret <7 x float> %c
@@ -333,602 +507,904 @@
 }
 
 define <4 x half> @min_v4f16(<4 x half> %a, <4 x half> %b) {
-; CHECK-NOFP16-LABEL: min_v4f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    // kill: def $d1 killed $d1 def $q1
-; CHECK-NOFP16-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NOFP16-NEXT:    mov h2, v1.h[1]
-; CHECK-NOFP16-NEXT:    mov h3, v0.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s4, h1
-; CHECK-NOFP16-NEXT:    fcvt s5, h0
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[2]
-; CHECK-NOFP16-NEXT:    mov h7, v0.h[2]
-; CHECK-NOFP16-NEXT:    mov h1, v1.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s2, h2
-; CHECK-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-NOFP16-NEXT:    fminnm s4, s5, s4
-; CHECK-NOFP16-NEXT:    fcvt s5, h7
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fminnm s2, s3, s2
-; CHECK-NOFP16-NEXT:    fcvt s3, h6
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[3]
-; CHECK-NOFP16-NEXT:    fcvt h0, s4
-; CHECK-NOFP16-NEXT:    fcvt h2, s2
-; CHECK-NOFP16-NEXT:    fminnm s3, s5, s3
-; CHECK-NOFP16-NEXT:    fcvt s4, h6
-; CHECK-NOFP16-NEXT:    mov v0.h[1], v2.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h2, s3
-; CHECK-NOFP16-NEXT:    fminnm s1, s4, s1
-; CHECK-NOFP16-NEXT:    mov v0.h[2], v2.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h1, s1
-; CHECK-NOFP16-NEXT:    mov v0.h[3], v1.h[0]
-; CHECK-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: min_v4f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fminnm v0.4h, v0.4h, v1.4h
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: min_v4f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
+; CHECK-NOFP16-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-NOFP16-SD-NEXT:    mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h1, v1.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT:    fminnm s4, s5, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h7
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fminnm s2, s3, s2
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h6
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s2
+; CHECK-NOFP16-SD-NEXT:    fminnm s3, s5, s3
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h6
+; CHECK-NOFP16-SD-NEXT:    mov v0.h[1], v2.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s3
+; CHECK-NOFP16-SD-NEXT:    fminnm s1, s4, s1
+; CHECK-NOFP16-SD-NEXT:    mov v0.h[2], v2.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h1, s1
+; CHECK-NOFP16-SD-NEXT:    mov v0.h[3], v1.h[0]
+; CHECK-NOFP16-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: min_v4f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fminnm v0.4h, v0.4h, v1.4h
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: min_v4f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v1.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT:    fminnm v0.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn v0.4h, v0.4s
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: min_v4f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    fminnm v0.4h, v0.4h, v1.4h
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call <4 x half> @llvm.minnum.v4f16(<4 x half> %a, <4 x half> %b)
   ret <4 x half> %c
 }
 
 define <4 x half> @max_v4f16(<4 x half> %a, <4 x half> %b) {
-; CHECK-NOFP16-LABEL: max_v4f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    // kill: def $d1 killed $d1 def $q1
-; CHECK-NOFP16-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-NOFP16-NEXT:    mov h2, v1.h[1]
-; CHECK-NOFP16-NEXT:    mov h3, v0.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s4, h1
-; CHECK-NOFP16-NEXT:    fcvt s5, h0
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[2]
-; CHECK-NOFP16-NEXT:    mov h7, v0.h[2]
-; CHECK-NOFP16-NEXT:    mov h1, v1.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s2, h2
-; CHECK-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-NOFP16-NEXT:    fmaxnm s4, s5, s4
-; CHECK-NOFP16-NEXT:    fcvt s5, h7
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fmaxnm s2, s3, s2
-; CHECK-NOFP16-NEXT:    fcvt s3, h6
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[3]
-; CHECK-NOFP16-NEXT:    fcvt h0, s4
-; CHECK-NOFP16-NEXT:    fcvt h2, s2
-; CHECK-NOFP16-NEXT:    fmaxnm s3, s5, s3
-; CHECK-NOFP16-NEXT:    fcvt s4, h6
-; CHECK-NOFP16-NEXT:    mov v0.h[1], v2.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h2, s3
-; CHECK-NOFP16-NEXT:    fmaxnm s1, s4, s1
-; CHECK-NOFP16-NEXT:    mov v0.h[2], v2.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h1, s1
-; CHECK-NOFP16-NEXT:    mov v0.h[3], v1.h[0]
-; CHECK-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: max_v4f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fmaxnm v0.4h, v0.4h, v1.4h
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: max_v4f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
+; CHECK-NOFP16-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-NOFP16-SD-NEXT:    mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h1, v1.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s4, s5, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h7
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s2, s3, s2
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h6
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s2
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s3, s5, s3
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h6
+; CHECK-NOFP16-SD-NEXT:    mov v0.h[1], v2.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s3
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s1, s4, s1
+; CHECK-NOFP16-SD-NEXT:    mov v0.h[2], v2.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h1, s1
+; CHECK-NOFP16-SD-NEXT:    mov v0.h[3], v1.h[0]
+; CHECK-NOFP16-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: max_v4f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fmaxnm v0.4h, v0.4h, v1.4h
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: max_v4f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v1.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v0.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn v0.4h, v0.4s
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: max_v4f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    fmaxnm v0.4h, v0.4h, v1.4h
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call <4 x half> @llvm.maxnum.v4f16(<4 x half> %a, <4 x half> %b)
   ret <4 x half> %c
 }
 
 define <7 x half> @min_v7f16(<7 x half> %a, <7 x half> %b) {
-; CHECK-NOFP16-LABEL: min_v7f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    mov h2, v1.h[1]
-; CHECK-NOFP16-NEXT:    mov h3, v0.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s4, h1
-; CHECK-NOFP16-NEXT:    fcvt s5, h0
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[2]
-; CHECK-NOFP16-NEXT:    mov h7, v0.h[2]
-; CHECK-NOFP16-NEXT:    mov h16, v1.h[3]
-; CHECK-NOFP16-NEXT:    mov h17, v0.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s2, h2
-; CHECK-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-NOFP16-NEXT:    fminnm s4, s5, s4
-; CHECK-NOFP16-NEXT:    fcvt s5, h6
-; CHECK-NOFP16-NEXT:    fcvt s6, h7
-; CHECK-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    fminnm s3, s3, s2
-; CHECK-NOFP16-NEXT:    fcvt h2, s4
-; CHECK-NOFP16-NEXT:    fminnm s4, s6, s5
-; CHECK-NOFP16-NEXT:    mov h5, v1.h[4]
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[4]
-; CHECK-NOFP16-NEXT:    fminnm s7, s16, s7
-; CHECK-NOFP16-NEXT:    fcvt h3, s3
-; CHECK-NOFP16-NEXT:    mov h16, v0.h[5]
-; CHECK-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-NOFP16-NEXT:    fcvt h7, s7
-; CHECK-NOFP16-NEXT:    mov v2.h[1], v3.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s4
-; CHECK-NOFP16-NEXT:    mov h4, v1.h[5]
-; CHECK-NOFP16-NEXT:    fminnm s5, s6, s5
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[6]
-; CHECK-NOFP16-NEXT:    mov v2.h[2], v3.h[0]
-; CHECK-NOFP16-NEXT:    mov h1, v1.h[7]
-; CHECK-NOFP16-NEXT:    fcvt s3, h4
-; CHECK-NOFP16-NEXT:    fcvt s4, h16
-; CHECK-NOFP16-NEXT:    mov h16, v0.h[6]
-; CHECK-NOFP16-NEXT:    mov h0, v0.h[7]
-; CHECK-NOFP16-NEXT:    mov v2.h[3], v7.h[0]
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fminnm s3, s4, s3
-; CHECK-NOFP16-NEXT:    fcvt h4, s5
-; CHECK-NOFP16-NEXT:    fcvt s5, h6
-; CHECK-NOFP16-NEXT:    fcvt s6, h16
-; CHECK-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-NOFP16-NEXT:    mov v2.h[4], v4.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s3
-; CHECK-NOFP16-NEXT:    fminnm s4, s6, s5
-; CHECK-NOFP16-NEXT:    fminnm s0, s0, s1
-; CHECK-NOFP16-NEXT:    mov v2.h[5], v3.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s4
-; CHECK-NOFP16-NEXT:    fcvt h0, s0
-; CHECK-NOFP16-NEXT:    mov v2.h[6], v3.h[0]
-; CHECK-NOFP16-NEXT:    mov v2.h[7], v0.h[0]
-; CHECK-NOFP16-NEXT:    mov v0.16b, v2.16b
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: min_v7f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fminnm v0.8h, v0.8h, v1.8h
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: min_v7f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h16, v1.h[3]
+; CHECK-NOFP16-SD-NEXT:    mov h17, v0.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT:    fminnm s4, s5, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h7
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    fminnm s3, s3, s2
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s4
+; CHECK-NOFP16-SD-NEXT:    fminnm s4, s6, s5
+; CHECK-NOFP16-SD-NEXT:    mov h5, v1.h[4]
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[4]
+; CHECK-NOFP16-SD-NEXT:    fminnm s7, s16, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT:    mov h16, v0.h[5]
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT:    mov h4, v1.h[5]
+; CHECK-NOFP16-SD-NEXT:    fminnm s5, s6, s5
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[2], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h4
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h16
+; CHECK-NOFP16-SD-NEXT:    mov h16, v0.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[3], v7.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fminnm s3, s4, s3
+; CHECK-NOFP16-SD-NEXT:    fcvt h4, s5
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[4], v4.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT:    fminnm s4, s6, s5
+; CHECK-NOFP16-SD-NEXT:    fminnm s0, s0, s1
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[5], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[6], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v0.16b, v2.16b
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: min_v7f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fminnm v0.8h, v0.8h, v1.8h
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: min_v7f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    mov h2, v0.h[4]
+; CHECK-NOFP16-GI-NEXT:    mov h3, v0.h[5]
+; CHECK-NOFP16-GI-NEXT:    mov h4, v1.h[4]
+; CHECK-NOFP16-GI-NEXT:    mov h5, v1.h[5]
+; CHECK-NOFP16-GI-NEXT:    mov h6, v0.h[6]
+; CHECK-NOFP16-GI-NEXT:    mov h7, v1.h[6]
+; CHECK-NOFP16-GI-NEXT:    fcvtl v1.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT:    mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v4.h[1], v5.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.h[2], v6.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v4.h[2], v7.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.h[3], v0.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v4.h[3], v0.h[0]
+; CHECK-NOFP16-GI-NEXT:    fcvtl v2.4s, v2.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v3.4s, v4.4h
+; CHECK-NOFP16-GI-NEXT:    mov s4, v2.s[1]
+; CHECK-NOFP16-GI-NEXT:    mov s6, v2.s[2]
+; CHECK-NOFP16-GI-NEXT:    mov s5, v3.s[1]
+; CHECK-NOFP16-GI-NEXT:    mov s7, v3.s[2]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[1], v4.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v3.s[1], v5.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[2], v6.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v3.s[2], v7.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v3.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT:    fminnm v2.4s, v2.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT:    fminnm v0.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT:    mov s1, v2.s[1]
+; CHECK-NOFP16-GI-NEXT:    mov s3, v2.s[2]
+; CHECK-NOFP16-GI-NEXT:    fcvtn v0.4h, v0.4s
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[1], v1.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov h1, v0.h[1]
+; CHECK-NOFP16-GI-NEXT:    mov h4, v0.h[2]
+; CHECK-NOFP16-GI-NEXT:    mov h5, v0.h[3]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[2], v3.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[2], v4.h[0]
+; CHECK-NOFP16-GI-NEXT:    fcvtn v1.4h, v2.4s
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[3], v5.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov h2, v1.h[1]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[4], v1.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov h1, v1.h[2]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[5], v2.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[6], v1.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[7], v0.h[0]
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: min_v7f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    mov h2, v0.h[1]
+; CHECK-FP16-GI-NEXT:    mov h3, v0.h[2]
+; CHECK-FP16-GI-NEXT:    mov h4, v0.h[3]
+; CHECK-FP16-GI-NEXT:    mov h5, v0.h[4]
+; CHECK-FP16-GI-NEXT:    mov h6, v0.h[5]
+; CHECK-FP16-GI-NEXT:    mov h7, v0.h[6]
+; CHECK-FP16-GI-NEXT:    mov h16, v1.h[1]
+; CHECK-FP16-GI-NEXT:    mov h17, v1.h[2]
+; CHECK-FP16-GI-NEXT:    mov v0.h[1], v2.h[0]
+; CHECK-FP16-GI-NEXT:    mov h18, v1.h[3]
+; CHECK-FP16-GI-NEXT:    mov h19, v1.h[4]
+; CHECK-FP16-GI-NEXT:    mov h20, v1.h[5]
+; CHECK-FP16-GI-NEXT:    mov h21, v1.h[6]
+; CHECK-FP16-GI-NEXT:    mov v1.h[1], v16.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[2], v3.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[2], v17.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[3], v4.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[3], v18.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[4], v5.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[4], v19.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[5], v6.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[5], v20.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[6], v7.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[6], v21.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT:    fminnm v0.8h, v0.8h, v1.8h
+; CHECK-FP16-GI-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-GI-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-GI-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-GI-NEXT:    mov h4, v0.h[4]
+; CHECK-FP16-GI-NEXT:    mov h5, v0.h[5]
+; CHECK-FP16-GI-NEXT:    mov h6, v0.h[6]
+; CHECK-FP16-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[2], v2.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[3], v3.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[4], v4.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[5], v5.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[6], v6.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call <7 x half> @llvm.minnum.v7f16(<7 x half> %a, <7 x half> %b)
   ret <7 x half> %c
 }
 
 define <7 x half> @max_v7f16(<7 x half> %a, <7 x half> %b) {
-; CHECK-NOFP16-LABEL: max_v7f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    mov h2, v1.h[1]
-; CHECK-NOFP16-NEXT:    mov h3, v0.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s4, h1
-; CHECK-NOFP16-NEXT:    fcvt s5, h0
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[2]
-; CHECK-NOFP16-NEXT:    mov h7, v0.h[2]
-; CHECK-NOFP16-NEXT:    mov h16, v1.h[3]
-; CHECK-NOFP16-NEXT:    mov h17, v0.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s2, h2
-; CHECK-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-NOFP16-NEXT:    fmaxnm s4, s5, s4
-; CHECK-NOFP16-NEXT:    fcvt s5, h6
-; CHECK-NOFP16-NEXT:    fcvt s6, h7
-; CHECK-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    fmaxnm s3, s3, s2
-; CHECK-NOFP16-NEXT:    fcvt h2, s4
-; CHECK-NOFP16-NEXT:    fmaxnm s4, s6, s5
-; CHECK-NOFP16-NEXT:    mov h5, v1.h[4]
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[4]
-; CHECK-NOFP16-NEXT:    fmaxnm s7, s16, s7
-; CHECK-NOFP16-NEXT:    fcvt h3, s3
-; CHECK-NOFP16-NEXT:    mov h16, v0.h[5]
-; CHECK-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-NOFP16-NEXT:    fcvt h7, s7
-; CHECK-NOFP16-NEXT:    mov v2.h[1], v3.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s4
-; CHECK-NOFP16-NEXT:    mov h4, v1.h[5]
-; CHECK-NOFP16-NEXT:    fmaxnm s5, s6, s5
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[6]
-; CHECK-NOFP16-NEXT:    mov v2.h[2], v3.h[0]
-; CHECK-NOFP16-NEXT:    mov h1, v1.h[7]
-; CHECK-NOFP16-NEXT:    fcvt s3, h4
-; CHECK-NOFP16-NEXT:    fcvt s4, h16
-; CHECK-NOFP16-NEXT:    mov h16, v0.h[6]
-; CHECK-NOFP16-NEXT:    mov h0, v0.h[7]
-; CHECK-NOFP16-NEXT:    mov v2.h[3], v7.h[0]
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fmaxnm s3, s4, s3
-; CHECK-NOFP16-NEXT:    fcvt h4, s5
-; CHECK-NOFP16-NEXT:    fcvt s5, h6
-; CHECK-NOFP16-NEXT:    fcvt s6, h16
-; CHECK-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-NOFP16-NEXT:    mov v2.h[4], v4.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s3
-; CHECK-NOFP16-NEXT:    fmaxnm s4, s6, s5
-; CHECK-NOFP16-NEXT:    fmaxnm s0, s0, s1
-; CHECK-NOFP16-NEXT:    mov v2.h[5], v3.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s4
-; CHECK-NOFP16-NEXT:    fcvt h0, s0
-; CHECK-NOFP16-NEXT:    mov v2.h[6], v3.h[0]
-; CHECK-NOFP16-NEXT:    mov v2.h[7], v0.h[0]
-; CHECK-NOFP16-NEXT:    mov v0.16b, v2.16b
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: max_v7f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fmaxnm v0.8h, v0.8h, v1.8h
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: max_v7f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h16, v1.h[3]
+; CHECK-NOFP16-SD-NEXT:    mov h17, v0.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s4, s5, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h7
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s3, s3, s2
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s4
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s4, s6, s5
+; CHECK-NOFP16-SD-NEXT:    mov h5, v1.h[4]
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[4]
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s7, s16, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT:    mov h16, v0.h[5]
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT:    mov h4, v1.h[5]
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s5, s6, s5
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[2], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h4
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h16
+; CHECK-NOFP16-SD-NEXT:    mov h16, v0.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[3], v7.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s3, s4, s3
+; CHECK-NOFP16-SD-NEXT:    fcvt h4, s5
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[4], v4.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s4, s6, s5
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s0, s0, s1
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[5], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[6], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v0.16b, v2.16b
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: max_v7f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fmaxnm v0.8h, v0.8h, v1.8h
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: max_v7f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    mov h2, v0.h[4]
+; CHECK-NOFP16-GI-NEXT:    mov h3, v0.h[5]
+; CHECK-NOFP16-GI-NEXT:    mov h4, v1.h[4]
+; CHECK-NOFP16-GI-NEXT:    mov h5, v1.h[5]
+; CHECK-NOFP16-GI-NEXT:    mov h6, v0.h[6]
+; CHECK-NOFP16-GI-NEXT:    mov h7, v1.h[6]
+; CHECK-NOFP16-GI-NEXT:    fcvtl v1.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT:    mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v4.h[1], v5.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.h[2], v6.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v4.h[2], v7.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.h[3], v0.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v4.h[3], v0.h[0]
+; CHECK-NOFP16-GI-NEXT:    fcvtl v2.4s, v2.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v3.4s, v4.4h
+; CHECK-NOFP16-GI-NEXT:    mov s4, v2.s[1]
+; CHECK-NOFP16-GI-NEXT:    mov s6, v2.s[2]
+; CHECK-NOFP16-GI-NEXT:    mov s5, v3.s[1]
+; CHECK-NOFP16-GI-NEXT:    mov s7, v3.s[2]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[1], v4.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v3.s[1], v5.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[2], v6.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v3.s[2], v7.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v3.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT:    fcvtl v0.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v2.4s, v2.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v0.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT:    mov s1, v2.s[1]
+; CHECK-NOFP16-GI-NEXT:    mov s3, v2.s[2]
+; CHECK-NOFP16-GI-NEXT:    fcvtn v0.4h, v0.4s
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[1], v1.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov h1, v0.h[1]
+; CHECK-NOFP16-GI-NEXT:    mov h4, v0.h[2]
+; CHECK-NOFP16-GI-NEXT:    mov h5, v0.h[3]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[2], v3.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v2.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[2], v4.h[0]
+; CHECK-NOFP16-GI-NEXT:    fcvtn v1.4h, v2.4s
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[3], v5.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov h2, v1.h[1]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[4], v1.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov h1, v1.h[2]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[5], v2.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[6], v1.h[0]
+; CHECK-NOFP16-GI-NEXT:    mov v0.h[7], v0.h[0]
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: max_v7f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    mov h2, v0.h[1]
+; CHECK-FP16-GI-NEXT:    mov h3, v0.h[2]
+; CHECK-FP16-GI-NEXT:    mov h4, v0.h[3]
+; CHECK-FP16-GI-NEXT:    mov h5, v0.h[4]
+; CHECK-FP16-GI-NEXT:    mov h6, v0.h[5]
+; CHECK-FP16-GI-NEXT:    mov h7, v0.h[6]
+; CHECK-FP16-GI-NEXT:    mov h16, v1.h[1]
+; CHECK-FP16-GI-NEXT:    mov h17, v1.h[2]
+; CHECK-FP16-GI-NEXT:    mov v0.h[1], v2.h[0]
+; CHECK-FP16-GI-NEXT:    mov h18, v1.h[3]
+; CHECK-FP16-GI-NEXT:    mov h19, v1.h[4]
+; CHECK-FP16-GI-NEXT:    mov h20, v1.h[5]
+; CHECK-FP16-GI-NEXT:    mov h21, v1.h[6]
+; CHECK-FP16-GI-NEXT:    mov v1.h[1], v16.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[2], v3.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[2], v17.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[3], v4.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[3], v18.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[4], v5.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[4], v19.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[5], v6.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[5], v20.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[6], v7.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[6], v21.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT:    mov v1.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT:    fmaxnm v0.8h, v0.8h, v1.8h
+; CHECK-FP16-GI-NEXT:    mov h1, v0.h[1]
+; CHECK-FP16-GI-NEXT:    mov h2, v0.h[2]
+; CHECK-FP16-GI-NEXT:    mov h3, v0.h[3]
+; CHECK-FP16-GI-NEXT:    mov h4, v0.h[4]
+; CHECK-FP16-GI-NEXT:    mov h5, v0.h[5]
+; CHECK-FP16-GI-NEXT:    mov h6, v0.h[6]
+; CHECK-FP16-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[2], v2.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[3], v3.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[4], v4.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[5], v5.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[6], v6.h[0]
+; CHECK-FP16-GI-NEXT:    mov v0.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call <7 x half> @llvm.maxnum.v7f16(<7 x half> %a, <7 x half> %b)
   ret <7 x half> %c
 }
 
 define <8 x half> @min_v8f16(<8 x half> %a, <8 x half> %b) {
-; CHECK-NOFP16-LABEL: min_v8f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    mov h2, v1.h[1]
-; CHECK-NOFP16-NEXT:    mov h3, v0.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s4, h1
-; CHECK-NOFP16-NEXT:    fcvt s5, h0
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[2]
-; CHECK-NOFP16-NEXT:    mov h7, v0.h[2]
-; CHECK-NOFP16-NEXT:    mov h16, v1.h[3]
-; CHECK-NOFP16-NEXT:    mov h17, v0.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s2, h2
-; CHECK-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-NOFP16-NEXT:    fminnm s4, s5, s4
-; CHECK-NOFP16-NEXT:    fcvt s5, h6
-; CHECK-NOFP16-NEXT:    fcvt s6, h7
-; CHECK-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    fminnm s3, s3, s2
-; CHECK-NOFP16-NEXT:    fcvt h2, s4
-; CHECK-NOFP16-NEXT:    fminnm s4, s6, s5
-; CHECK-NOFP16-NEXT:    mov h5, v1.h[4]
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[4]
-; CHECK-NOFP16-NEXT:    fminnm s7, s16, s7
-; CHECK-NOFP16-NEXT:    fcvt h3, s3
-; CHECK-NOFP16-NEXT:    mov h16, v0.h[5]
-; CHECK-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-NOFP16-NEXT:    fcvt h7, s7
-; CHECK-NOFP16-NEXT:    mov v2.h[1], v3.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s4
-; CHECK-NOFP16-NEXT:    mov h4, v1.h[5]
-; CHECK-NOFP16-NEXT:    fminnm s5, s6, s5
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[6]
-; CHECK-NOFP16-NEXT:    mov v2.h[2], v3.h[0]
-; CHECK-NOFP16-NEXT:    mov h1, v1.h[7]
-; CHECK-NOFP16-NEXT:    fcvt s3, h4
-; CHECK-NOFP16-NEXT:    fcvt s4, h16
-; CHECK-NOFP16-NEXT:    mov h16, v0.h[6]
-; CHECK-NOFP16-NEXT:    mov h0, v0.h[7]
-; CHECK-NOFP16-NEXT:    mov v2.h[3], v7.h[0]
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fminnm s3, s4, s3
-; CHECK-NOFP16-NEXT:    fcvt h4, s5
-; CHECK-NOFP16-NEXT:    fcvt s5, h6
-; CHECK-NOFP16-NEXT:    fcvt s6, h16
-; CHECK-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-NOFP16-NEXT:    mov v2.h[4], v4.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s3
-; CHECK-NOFP16-NEXT:    fminnm s4, s6, s5
-; CHECK-NOFP16-NEXT:    fminnm s0, s0, s1
-; CHECK-NOFP16-NEXT:    mov v2.h[5], v3.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s4
-; CHECK-NOFP16-NEXT:    fcvt h0, s0
-; CHECK-NOFP16-NEXT:    mov v2.h[6], v3.h[0]
-; CHECK-NOFP16-NEXT:    mov v2.h[7], v0.h[0]
-; CHECK-NOFP16-NEXT:    mov v0.16b, v2.16b
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: min_v8f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fminnm v0.8h, v0.8h, v1.8h
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: min_v8f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h16, v1.h[3]
+; CHECK-NOFP16-SD-NEXT:    mov h17, v0.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT:    fminnm s4, s5, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h7
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    fminnm s3, s3, s2
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s4
+; CHECK-NOFP16-SD-NEXT:    fminnm s4, s6, s5
+; CHECK-NOFP16-SD-NEXT:    mov h5, v1.h[4]
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[4]
+; CHECK-NOFP16-SD-NEXT:    fminnm s7, s16, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT:    mov h16, v0.h[5]
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT:    mov h4, v1.h[5]
+; CHECK-NOFP16-SD-NEXT:    fminnm s5, s6, s5
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[2], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h4
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h16
+; CHECK-NOFP16-SD-NEXT:    mov h16, v0.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[3], v7.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fminnm s3, s4, s3
+; CHECK-NOFP16-SD-NEXT:    fcvt h4, s5
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[4], v4.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT:    fminnm s4, s6, s5
+; CHECK-NOFP16-SD-NEXT:    fminnm s0, s0, s1
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[5], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[6], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v0.16b, v2.16b
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: min_v8f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fminnm v0.8h, v0.8h, v1.8h
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: min_v8f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    fcvtl v2.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v3.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v1.4s, v1.8h
+; CHECK-NOFP16-GI-NEXT:    fminnm v2.4s, v2.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT:    fminnm v1.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn v0.4h, v2.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn2 v0.8h, v1.4s
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: min_v8f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    fminnm v0.8h, v0.8h, v1.8h
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call <8 x half> @llvm.minnum.v8f16(<8 x half> %a, <8 x half> %b)
   ret <8 x half> %c
 }
 
 define <8 x half> @max_v8f16(<8 x half> %a, <8 x half> %b) {
-; CHECK-NOFP16-LABEL: max_v8f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    mov h2, v1.h[1]
-; CHECK-NOFP16-NEXT:    mov h3, v0.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s4, h1
-; CHECK-NOFP16-NEXT:    fcvt s5, h0
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[2]
-; CHECK-NOFP16-NEXT:    mov h7, v0.h[2]
-; CHECK-NOFP16-NEXT:    mov h16, v1.h[3]
-; CHECK-NOFP16-NEXT:    mov h17, v0.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s2, h2
-; CHECK-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-NOFP16-NEXT:    fmaxnm s4, s5, s4
-; CHECK-NOFP16-NEXT:    fcvt s5, h6
-; CHECK-NOFP16-NEXT:    fcvt s6, h7
-; CHECK-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    fmaxnm s3, s3, s2
-; CHECK-NOFP16-NEXT:    fcvt h2, s4
-; CHECK-NOFP16-NEXT:    fmaxnm s4, s6, s5
-; CHECK-NOFP16-NEXT:    mov h5, v1.h[4]
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[4]
-; CHECK-NOFP16-NEXT:    fmaxnm s7, s16, s7
-; CHECK-NOFP16-NEXT:    fcvt h3, s3
-; CHECK-NOFP16-NEXT:    mov h16, v0.h[5]
-; CHECK-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-NOFP16-NEXT:    fcvt h7, s7
-; CHECK-NOFP16-NEXT:    mov v2.h[1], v3.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s4
-; CHECK-NOFP16-NEXT:    mov h4, v1.h[5]
-; CHECK-NOFP16-NEXT:    fmaxnm s5, s6, s5
-; CHECK-NOFP16-NEXT:    mov h6, v1.h[6]
-; CHECK-NOFP16-NEXT:    mov v2.h[2], v3.h[0]
-; CHECK-NOFP16-NEXT:    mov h1, v1.h[7]
-; CHECK-NOFP16-NEXT:    fcvt s3, h4
-; CHECK-NOFP16-NEXT:    fcvt s4, h16
-; CHECK-NOFP16-NEXT:    mov h16, v0.h[6]
-; CHECK-NOFP16-NEXT:    mov h0, v0.h[7]
-; CHECK-NOFP16-NEXT:    mov v2.h[3], v7.h[0]
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fmaxnm s3, s4, s3
-; CHECK-NOFP16-NEXT:    fcvt h4, s5
-; CHECK-NOFP16-NEXT:    fcvt s5, h6
-; CHECK-NOFP16-NEXT:    fcvt s6, h16
-; CHECK-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-NOFP16-NEXT:    mov v2.h[4], v4.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s3
-; CHECK-NOFP16-NEXT:    fmaxnm s4, s6, s5
-; CHECK-NOFP16-NEXT:    fmaxnm s0, s0, s1
-; CHECK-NOFP16-NEXT:    mov v2.h[5], v3.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h3, s4
-; CHECK-NOFP16-NEXT:    fcvt h0, s0
-; CHECK-NOFP16-NEXT:    mov v2.h[6], v3.h[0]
-; CHECK-NOFP16-NEXT:    mov v2.h[7], v0.h[0]
-; CHECK-NOFP16-NEXT:    mov v0.16b, v2.16b
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: max_v8f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fmaxnm v0.8h, v0.8h, v1.8h
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: max_v8f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h16, v1.h[3]
+; CHECK-NOFP16-SD-NEXT:    mov h17, v0.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s4, s5, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h7
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s3, s3, s2
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s4
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s4, s6, s5
+; CHECK-NOFP16-SD-NEXT:    mov h5, v1.h[4]
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[4]
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s7, s16, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT:    mov h16, v0.h[5]
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT:    mov h4, v1.h[5]
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s5, s6, s5
+; CHECK-NOFP16-SD-NEXT:    mov h6, v1.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[2], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h4
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h16
+; CHECK-NOFP16-SD-NEXT:    mov h16, v0.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[3], v7.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s3, s4, s3
+; CHECK-NOFP16-SD-NEXT:    fcvt h4, s5
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[4], v4.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s4, s6, s5
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s0, s0, s1
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[5], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[6], v3.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v2.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v0.16b, v2.16b
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: max_v8f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fmaxnm v0.8h, v0.8h, v1.8h
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: max_v8f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    fcvtl v2.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v3.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v1.4s, v1.8h
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v2.4s, v2.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v1.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn v0.4h, v2.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn2 v0.8h, v1.4s
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: max_v8f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    fmaxnm v0.8h, v0.8h, v1.8h
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call <8 x half> @llvm.maxnum.v8f16(<8 x half> %a, <8 x half> %b)
   ret <8 x half> %c
 }
 
 define <16 x half> @min_v16f16(<16 x half> %a, <16 x half> %b) {
-; CHECK-NOFP16-LABEL: min_v16f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    mov h4, v2.h[1]
-; CHECK-NOFP16-NEXT:    mov h5, v0.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s6, h2
-; CHECK-NOFP16-NEXT:    fcvt s7, h0
-; CHECK-NOFP16-NEXT:    mov h16, v2.h[2]
-; CHECK-NOFP16-NEXT:    mov h17, v0.h[2]
-; CHECK-NOFP16-NEXT:    mov h18, v3.h[1]
-; CHECK-NOFP16-NEXT:    mov h19, v1.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s4, h4
-; CHECK-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-NOFP16-NEXT:    fminnm s6, s7, s6
-; CHECK-NOFP16-NEXT:    fcvt s20, h1
-; CHECK-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    mov h17, v2.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s18, h18
-; CHECK-NOFP16-NEXT:    fminnm s5, s5, s4
-; CHECK-NOFP16-NEXT:    fcvt s19, h19
-; CHECK-NOFP16-NEXT:    fcvt h4, s6
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[3]
-; CHECK-NOFP16-NEXT:    fminnm s7, s16, s7
-; CHECK-NOFP16-NEXT:    fcvt s16, h3
-; CHECK-NOFP16-NEXT:    mov h21, v0.h[4]
-; CHECK-NOFP16-NEXT:    fcvt s17, h17
-; CHECK-NOFP16-NEXT:    fcvt h5, s5
-; CHECK-NOFP16-NEXT:    fminnm s18, s19, s18
-; CHECK-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-NOFP16-NEXT:    mov h19, v3.h[2]
-; CHECK-NOFP16-NEXT:    fminnm s16, s20, s16
-; CHECK-NOFP16-NEXT:    mov h20, v1.h[2]
-; CHECK-NOFP16-NEXT:    fcvt s21, h21
-; CHECK-NOFP16-NEXT:    mov v4.h[1], v5.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h5, s7
-; CHECK-NOFP16-NEXT:    mov h7, v2.h[4]
-; CHECK-NOFP16-NEXT:    fminnm s6, s6, s17
-; CHECK-NOFP16-NEXT:    mov h17, v3.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s19, h19
-; CHECK-NOFP16-NEXT:    fcvt s20, h20
-; CHECK-NOFP16-NEXT:    mov v4.h[2], v5.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h5, s16
-; CHECK-NOFP16-NEXT:    fcvt s7, h7
-; CHECK-NOFP16-NEXT:    fcvt h16, s18
-; CHECK-NOFP16-NEXT:    mov h18, v1.h[3]
-; CHECK-NOFP16-NEXT:    fcvt h6, s6
-; CHECK-NOFP16-NEXT:    fminnm s7, s21, s7
-; CHECK-NOFP16-NEXT:    mov v5.h[1], v16.h[0]
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    fminnm s17, s20, s19
-; CHECK-NOFP16-NEXT:    fcvt s18, h18
-; CHECK-NOFP16-NEXT:    mov h19, v3.h[4]
-; CHECK-NOFP16-NEXT:    mov h20, v1.h[4]
-; CHECK-NOFP16-NEXT:    mov v4.h[3], v6.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h6, s7
-; CHECK-NOFP16-NEXT:    fminnm s7, s18, s16
-; CHECK-NOFP16-NEXT:    fcvt h16, s17
-; CHECK-NOFP16-NEXT:    fcvt s17, h19
-; CHECK-NOFP16-NEXT:    fcvt s18, h20
-; CHECK-NOFP16-NEXT:    mov v4.h[4], v6.h[0]
-; CHECK-NOFP16-NEXT:    mov h19, v1.h[5]
-; CHECK-NOFP16-NEXT:    mov v5.h[2], v16.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h6, s7
-; CHECK-NOFP16-NEXT:    fminnm s7, s18, s17
-; CHECK-NOFP16-NEXT:    mov h16, v2.h[5]
-; CHECK-NOFP16-NEXT:    mov h17, v0.h[5]
-; CHECK-NOFP16-NEXT:    mov h18, v3.h[5]
-; CHECK-NOFP16-NEXT:    mov v5.h[3], v6.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h6, s7
-; CHECK-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    fcvt s17, h18
-; CHECK-NOFP16-NEXT:    fcvt s18, h19
-; CHECK-NOFP16-NEXT:    mov h19, v2.h[6]
-; CHECK-NOFP16-NEXT:    mov h2, v2.h[7]
-; CHECK-NOFP16-NEXT:    mov v5.h[4], v6.h[0]
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[6]
-; CHECK-NOFP16-NEXT:    fminnm s7, s16, s7
-; CHECK-NOFP16-NEXT:    fminnm s16, s18, s17
-; CHECK-NOFP16-NEXT:    mov h17, v3.h[6]
-; CHECK-NOFP16-NEXT:    mov h18, v1.h[6]
-; CHECK-NOFP16-NEXT:    fcvt s19, h19
-; CHECK-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-NOFP16-NEXT:    mov h0, v0.h[7]
-; CHECK-NOFP16-NEXT:    mov h3, v3.h[7]
-; CHECK-NOFP16-NEXT:    mov h1, v1.h[7]
-; CHECK-NOFP16-NEXT:    fcvt s17, h17
-; CHECK-NOFP16-NEXT:    fcvt h7, s7
-; CHECK-NOFP16-NEXT:    fcvt s18, h18
-; CHECK-NOFP16-NEXT:    fcvt s2, h2
-; CHECK-NOFP16-NEXT:    fminnm s6, s6, s19
-; CHECK-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fcvt h16, s16
-; CHECK-NOFP16-NEXT:    fminnm s17, s18, s17
-; CHECK-NOFP16-NEXT:    mov v4.h[5], v7.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h6, s6
-; CHECK-NOFP16-NEXT:    fminnm s0, s0, s2
-; CHECK-NOFP16-NEXT:    fminnm s1, s1, s3
-; CHECK-NOFP16-NEXT:    mov v5.h[5], v16.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h2, s17
-; CHECK-NOFP16-NEXT:    mov v4.h[6], v6.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h0, s0
-; CHECK-NOFP16-NEXT:    fcvt h1, s1
-; CHECK-NOFP16-NEXT:    mov v5.h[6], v2.h[0]
-; CHECK-NOFP16-NEXT:    mov v4.h[7], v0.h[0]
-; CHECK-NOFP16-NEXT:    mov v5.h[7], v1.h[0]
-; CHECK-NOFP16-NEXT:    mov v0.16b, v4.16b
-; CHECK-NOFP16-NEXT:    mov v1.16b, v5.16b
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: min_v16f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fminnm v0.8h, v0.8h, v2.8h
-; CHECK-FP16-NEXT:    fminnm v1.8h, v1.8h, v3.8h
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: min_v16f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    mov h4, v2.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h5, v0.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h2
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h0
+; CHECK-NOFP16-SD-NEXT:    mov h16, v2.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h17, v0.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h18, v3.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h19, v1.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h4
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT:    fminnm s6, s7, s6
+; CHECK-NOFP16-SD-NEXT:    fcvt s20, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    mov h17, v2.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT:    fminnm s5, s5, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT:    fcvt h4, s6
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[3]
+; CHECK-NOFP16-SD-NEXT:    fminnm s7, s16, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h3
+; CHECK-NOFP16-SD-NEXT:    mov h21, v0.h[4]
+; CHECK-NOFP16-SD-NEXT:    fcvt s17, h17
+; CHECK-NOFP16-SD-NEXT:    fcvt h5, s5
+; CHECK-NOFP16-SD-NEXT:    fminnm s18, s19, s18
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT:    mov h19, v3.h[2]
+; CHECK-NOFP16-SD-NEXT:    fminnm s16, s20, s16
+; CHECK-NOFP16-SD-NEXT:    mov h20, v1.h[2]
+; CHECK-NOFP16-SD-NEXT:    fcvt s21, h21
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[1], v5.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h5, s7
+; CHECK-NOFP16-SD-NEXT:    mov h7, v2.h[4]
+; CHECK-NOFP16-SD-NEXT:    fminnm s6, s6, s17
+; CHECK-NOFP16-SD-NEXT:    mov h17, v3.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT:    fcvt s20, h20
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[2], v5.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h5, s16
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h7
+; CHECK-NOFP16-SD-NEXT:    fcvt h16, s18
+; CHECK-NOFP16-SD-NEXT:    mov h18, v1.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s6
+; CHECK-NOFP16-SD-NEXT:    fminnm s7, s21, s7
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[1], v16.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    fminnm s17, s20, s19
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT:    mov h19, v3.h[4]
+; CHECK-NOFP16-SD-NEXT:    mov h20, v1.h[4]
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[3], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT:    fminnm s7, s18, s16
+; CHECK-NOFP16-SD-NEXT:    fcvt h16, s17
+; CHECK-NOFP16-SD-NEXT:    fcvt s17, h19
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h20
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[4], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov h19, v1.h[5]
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[2], v16.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT:    fminnm s7, s18, s17
+; CHECK-NOFP16-SD-NEXT:    mov h16, v2.h[5]
+; CHECK-NOFP16-SD-NEXT:    mov h17, v0.h[5]
+; CHECK-NOFP16-SD-NEXT:    mov h18, v3.h[5]
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[3], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    fcvt s17, h18
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h19
+; CHECK-NOFP16-SD-NEXT:    mov h19, v2.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov h2, v2.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[4], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[6]
+; CHECK-NOFP16-SD-NEXT:    fminnm s7, s16, s7
+; CHECK-NOFP16-SD-NEXT:    fminnm s16, s18, s17
+; CHECK-NOFP16-SD-NEXT:    mov h17, v3.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov h18, v1.h[6]
+; CHECK-NOFP16-SD-NEXT:    fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT:    mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov h3, v3.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT:    fcvt s17, h17
+; CHECK-NOFP16-SD-NEXT:    fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT:    fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT:    fminnm s6, s6, s19
+; CHECK-NOFP16-SD-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt h16, s16
+; CHECK-NOFP16-SD-NEXT:    fminnm s17, s18, s17
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[5], v7.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s6
+; CHECK-NOFP16-SD-NEXT:    fminnm s0, s0, s2
+; CHECK-NOFP16-SD-NEXT:    fminnm s1, s1, s3
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[5], v16.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s17
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[6], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT:    fcvt h1, s1
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[6], v2.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[7], v1.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v0.16b, v4.16b
+; CHECK-NOFP16-SD-NEXT:    mov v1.16b, v5.16b
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: min_v16f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fminnm v0.8h, v0.8h, v2.8h
+; CHECK-FP16-SD-NEXT:    fminnm v1.8h, v1.8h, v3.8h
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: min_v16f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    fcvtl v4.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v5.4s, v2.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v6.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v7.4s, v3.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v2.4s, v2.8h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v1.4s, v1.8h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v3.4s, v3.8h
+; CHECK-NOFP16-GI-NEXT:    fminnm v4.4s, v4.4s, v5.4s
+; CHECK-NOFP16-GI-NEXT:    fminnm v5.4s, v6.4s, v7.4s
+; CHECK-NOFP16-GI-NEXT:    fminnm v2.4s, v0.4s, v2.4s
+; CHECK-NOFP16-GI-NEXT:    fminnm v3.4s, v1.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn v0.4h, v4.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn v1.4h, v5.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn2 v0.8h, v2.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn2 v1.8h, v3.4s
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: min_v16f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    fminnm v0.8h, v0.8h, v2.8h
+; CHECK-FP16-GI-NEXT:    fminnm v1.8h, v1.8h, v3.8h
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call <16 x half> @llvm.minnum.v16f16(<16 x half> %a, <16 x half> %b)
   ret <16 x half> %c
 }
 
 define <16 x half> @max_v16f16(<16 x half> %a, <16 x half> %b) {
-; CHECK-NOFP16-LABEL: max_v16f16:
-; CHECK-NOFP16:       // %bb.0: // %entry
-; CHECK-NOFP16-NEXT:    mov h4, v2.h[1]
-; CHECK-NOFP16-NEXT:    mov h5, v0.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s6, h2
-; CHECK-NOFP16-NEXT:    fcvt s7, h0
-; CHECK-NOFP16-NEXT:    mov h16, v2.h[2]
-; CHECK-NOFP16-NEXT:    mov h17, v0.h[2]
-; CHECK-NOFP16-NEXT:    mov h18, v3.h[1]
-; CHECK-NOFP16-NEXT:    mov h19, v1.h[1]
-; CHECK-NOFP16-NEXT:    fcvt s4, h4
-; CHECK-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-NOFP16-NEXT:    fmaxnm s6, s7, s6
-; CHECK-NOFP16-NEXT:    fcvt s20, h1
-; CHECK-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    mov h17, v2.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s18, h18
-; CHECK-NOFP16-NEXT:    fmaxnm s5, s5, s4
-; CHECK-NOFP16-NEXT:    fcvt s19, h19
-; CHECK-NOFP16-NEXT:    fcvt h4, s6
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[3]
-; CHECK-NOFP16-NEXT:    fmaxnm s7, s16, s7
-; CHECK-NOFP16-NEXT:    fcvt s16, h3
-; CHECK-NOFP16-NEXT:    mov h21, v0.h[4]
-; CHECK-NOFP16-NEXT:    fcvt s17, h17
-; CHECK-NOFP16-NEXT:    fcvt h5, s5
-; CHECK-NOFP16-NEXT:    fmaxnm s18, s19, s18
-; CHECK-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-NOFP16-NEXT:    mov h19, v3.h[2]
-; CHECK-NOFP16-NEXT:    fmaxnm s16, s20, s16
-; CHECK-NOFP16-NEXT:    mov h20, v1.h[2]
-; CHECK-NOFP16-NEXT:    fcvt s21, h21
-; CHECK-NOFP16-NEXT:    mov v4.h[1], v5.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h5, s7
-; CHECK-NOFP16-NEXT:    mov h7, v2.h[4]
-; CHECK-NOFP16-NEXT:    fmaxnm s6, s6, s17
-; CHECK-NOFP16-NEXT:    mov h17, v3.h[3]
-; CHECK-NOFP16-NEXT:    fcvt s19, h19
-; CHECK-NOFP16-NEXT:    fcvt s20, h20
-; CHECK-NOFP16-NEXT:    mov v4.h[2], v5.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h5, s16
-; CHECK-NOFP16-NEXT:    fcvt s7, h7
-; CHECK-NOFP16-NEXT:    fcvt h16, s18
-; CHECK-NOFP16-NEXT:    mov h18, v1.h[3]
-; CHECK-NOFP16-NEXT:    fcvt h6, s6
-; CHECK-NOFP16-NEXT:    fmaxnm s7, s21, s7
-; CHECK-NOFP16-NEXT:    mov v5.h[1], v16.h[0]
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    fmaxnm s17, s20, s19
-; CHECK-NOFP16-NEXT:    fcvt s18, h18
-; CHECK-NOFP16-NEXT:    mov h19, v3.h[4]
-; CHECK-NOFP16-NEXT:    mov h20, v1.h[4]
-; CHECK-NOFP16-NEXT:    mov v4.h[3], v6.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h6, s7
-; CHECK-NOFP16-NEXT:    fmaxnm s7, s18, s16
-; CHECK-NOFP16-NEXT:    fcvt h16, s17
-; CHECK-NOFP16-NEXT:    fcvt s17, h19
-; CHECK-NOFP16-NEXT:    fcvt s18, h20
-; CHECK-NOFP16-NEXT:    mov v4.h[4], v6.h[0]
-; CHECK-NOFP16-NEXT:    mov h19, v1.h[5]
-; CHECK-NOFP16-NEXT:    mov v5.h[2], v16.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h6, s7
-; CHECK-NOFP16-NEXT:    fmaxnm s7, s18, s17
-; CHECK-NOFP16-NEXT:    mov h16, v2.h[5]
-; CHECK-NOFP16-NEXT:    mov h17, v0.h[5]
-; CHECK-NOFP16-NEXT:    mov h18, v3.h[5]
-; CHECK-NOFP16-NEXT:    mov v5.h[3], v6.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h6, s7
-; CHECK-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-NOFP16-NEXT:    fcvt s16, h17
-; CHECK-NOFP16-NEXT:    fcvt s17, h18
-; CHECK-NOFP16-NEXT:    fcvt s18, h19
-; CHECK-NOFP16-NEXT:    mov h19, v2.h[6]
-; CHECK-NOFP16-NEXT:    mov h2, v2.h[7]
-; CHECK-NOFP16-NEXT:    mov v5.h[4], v6.h[0]
-; CHECK-NOFP16-NEXT:    mov h6, v0.h[6]
-; CHECK-NOFP16-NEXT:    fmaxnm s7, s16, s7
-; CHECK-NOFP16-NEXT:    fmaxnm s16, s18, s17
-; CHECK-NOFP16-NEXT:    mov h17, v3.h[6]
-; CHECK-NOFP16-NEXT:    mov h18, v1.h[6]
-; CHECK-NOFP16-NEXT:    fcvt s19, h19
-; CHECK-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-NOFP16-NEXT:    mov h0, v0.h[7]
-; CHECK-NOFP16-NEXT:    mov h3, v3.h[7]
-; CHECK-NOFP16-NEXT:    mov h1, v1.h[7]
-; CHECK-NOFP16-NEXT:    fcvt s17, h17
-; CHECK-NOFP16-NEXT:    fcvt h7, s7
-; CHECK-NOFP16-NEXT:    fcvt s18, h18
-; CHECK-NOFP16-NEXT:    fcvt s2, h2
-; CHECK-NOFP16-NEXT:    fmaxnm s6, s6, s19
-; CHECK-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-NOFP16-NEXT:    fcvt h16, s16
-; CHECK-NOFP16-NEXT:    fmaxnm s17, s18, s17
-; CHECK-NOFP16-NEXT:    mov v4.h[5], v7.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h6, s6
-; CHECK-NOFP16-NEXT:    fmaxnm s0, s0, s2
-; CHECK-NOFP16-NEXT:    fmaxnm s1, s1, s3
-; CHECK-NOFP16-NEXT:    mov v5.h[5], v16.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h2, s17
-; CHECK-NOFP16-NEXT:    mov v4.h[6], v6.h[0]
-; CHECK-NOFP16-NEXT:    fcvt h0, s0
-; CHECK-NOFP16-NEXT:    fcvt h1, s1
-; CHECK-NOFP16-NEXT:    mov v5.h[6], v2.h[0]
-; CHECK-NOFP16-NEXT:    mov v4.h[7], v0.h[0]
-; CHECK-NOFP16-NEXT:    mov v5.h[7], v1.h[0]
-; CHECK-NOFP16-NEXT:    mov v0.16b, v4.16b
-; CHECK-NOFP16-NEXT:    mov v1.16b, v5.16b
-; CHECK-NOFP16-NEXT:    ret
-;
-; CHECK-FP16-LABEL: max_v16f16:
-; CHECK-FP16:       // %bb.0: // %entry
-; CHECK-FP16-NEXT:    fmaxnm v0.8h, v0.8h, v2.8h
-; CHECK-FP16-NEXT:    fmaxnm v1.8h, v1.8h, v3.8h
-; CHECK-FP16-NEXT:    ret
+; CHECK-NOFP16-SD-LABEL: max_v16f16:
+; CHECK-NOFP16-SD:       // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT:    mov h4, v2.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h5, v0.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h2
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h0
+; CHECK-NOFP16-SD-NEXT:    mov h16, v2.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h17, v0.h[2]
+; CHECK-NOFP16-SD-NEXT:    mov h18, v3.h[1]
+; CHECK-NOFP16-SD-NEXT:    mov h19, v1.h[1]
+; CHECK-NOFP16-SD-NEXT:    fcvt s4, h4
+; CHECK-NOFP16-SD-NEXT:    fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s6, s7, s6
+; CHECK-NOFP16-SD-NEXT:    fcvt s20, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    mov h17, v2.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s5, s5, s4
+; CHECK-NOFP16-SD-NEXT:    fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT:    fcvt h4, s6
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[3]
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s7, s16, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h3
+; CHECK-NOFP16-SD-NEXT:    mov h21, v0.h[4]
+; CHECK-NOFP16-SD-NEXT:    fcvt s17, h17
+; CHECK-NOFP16-SD-NEXT:    fcvt h5, s5
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s18, s19, s18
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT:    mov h19, v3.h[2]
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s16, s20, s16
+; CHECK-NOFP16-SD-NEXT:    mov h20, v1.h[2]
+; CHECK-NOFP16-SD-NEXT:    fcvt s21, h21
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[1], v5.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h5, s7
+; CHECK-NOFP16-SD-NEXT:    mov h7, v2.h[4]
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s6, s6, s17
+; CHECK-NOFP16-SD-NEXT:    mov h17, v3.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT:    fcvt s20, h20
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[2], v5.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h5, s16
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h7
+; CHECK-NOFP16-SD-NEXT:    fcvt h16, s18
+; CHECK-NOFP16-SD-NEXT:    mov h18, v1.h[3]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s6
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s7, s21, s7
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[1], v16.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s17, s20, s19
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT:    mov h19, v3.h[4]
+; CHECK-NOFP16-SD-NEXT:    mov h20, v1.h[4]
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[3], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s7, s18, s16
+; CHECK-NOFP16-SD-NEXT:    fcvt h16, s17
+; CHECK-NOFP16-SD-NEXT:    fcvt s17, h19
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h20
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[4], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov h19, v1.h[5]
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[2], v16.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s7, s18, s17
+; CHECK-NOFP16-SD-NEXT:    mov h16, v2.h[5]
+; CHECK-NOFP16-SD-NEXT:    mov h17, v0.h[5]
+; CHECK-NOFP16-SD-NEXT:    mov h18, v3.h[5]
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[3], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT:    fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT:    fcvt s17, h18
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h19
+; CHECK-NOFP16-SD-NEXT:    mov h19, v2.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov h2, v2.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[4], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov h6, v0.h[6]
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s7, s16, s7
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s16, s18, s17
+; CHECK-NOFP16-SD-NEXT:    mov h17, v3.h[6]
+; CHECK-NOFP16-SD-NEXT:    mov h18, v1.h[6]
+; CHECK-NOFP16-SD-NEXT:    fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT:    fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT:    mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov h3, v3.h[7]
+; CHECK-NOFP16-SD-NEXT:    mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT:    fcvt s17, h17
+; CHECK-NOFP16-SD-NEXT:    fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT:    fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT:    fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s6, s6, s19
+; CHECK-NOFP16-SD-NEXT:    fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT:    fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT:    fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT:    fcvt h16, s16
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s17, s18, s17
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[5], v7.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h6, s6
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s0, s0, s2
+; CHECK-NOFP16-SD-NEXT:    fmaxnm s1, s1, s3
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[5], v16.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h2, s17
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[6], v6.h[0]
+; CHECK-NOFP16-SD-NEXT:    fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT:    fcvt h1, s1
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[6], v2.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v4.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v5.h[7], v1.h[0]
+; CHECK-NOFP16-SD-NEXT:    mov v0.16b, v4.16b
+; CHECK-NOFP16-SD-NEXT:    mov v1.16b, v5.16b
+; CHECK-NOFP16-SD-NEXT:    ret
+;
+; CHECK-FP16-SD-LABEL: max_v16f16:
+; CHECK-FP16-SD:       // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT:    fmaxnm v0.8h, v0.8h, v2.8h
+; CHECK-FP16-SD-NEXT:    fmaxnm v1.8h, v1.8h, v3.8h
+; CHECK-FP16-SD-NEXT:    ret
+;
+; CHECK-NOFP16-GI-LABEL: max_v16f16:
+; CHECK-NOFP16-GI:       // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT:    fcvtl v4.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v5.4s, v2.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v6.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl v7.4s, v3.4h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v0.4s, v0.8h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v2.4s, v2.8h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v1.4s, v1.8h
+; CHECK-NOFP16-GI-NEXT:    fcvtl2 v3.4s, v3.8h
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v4.4s, v4.4s, v5.4s
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v5.4s, v6.4s, v7.4s
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v2.4s, v0.4s, v2.4s
+; CHECK-NOFP16-GI-NEXT:    fmaxnm v3.4s, v1.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn v0.4h, v4.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn v1.4h, v5.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn2 v0.8h, v2.4s
+; CHECK-NOFP16-GI-NEXT:    fcvtn2 v1.8h, v3.4s
+; CHECK-NOFP16-GI-NEXT:    ret
+;
+; CHECK-FP16-GI-LABEL: max_v16f16:
+; CHECK-FP16-GI:       // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT:    fmaxnm v0.8h, v0.8h, v2.8h
+; CHECK-FP16-GI-NEXT:    fmaxnm v1.8h, v1.8h, v3.8h
+; CHECK-FP16-GI-NEXT:    ret
 entry:
   %c = call <16 x half> @llvm.maxnum.v16f16(<16 x half> %a, <16 x half> %b)
   ret <16 x half> %c