diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.h b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.h
index 72401a7a259c0..e0ce3d1dcb620 100644
--- a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.h
+++ b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.h
@@ -40,6 +40,8 @@ class WebAssemblyTargetLowering final : public TargetLowering {
   MVT getPointerTy(const DataLayout &DL, uint32_t AS = 0) const override;
   MVT getPointerMemTy(const DataLayout &DL, uint32_t AS = 0) const override;
 
+  bool softPromoteHalfType() const override { return true; }
+
 private:
   /// Keep a pointer to the WebAssemblySubtarget around so that we can make the
   /// right decision when generating code for different targets.
diff --git a/llvm/test/CodeGen/Generic/half.ll b/llvm/test/CodeGen/Generic/half.ll
index ef7bfe2f2d9ce..55831f9438df6 100644
--- a/llvm/test/CodeGen/Generic/half.ll
+++ b/llvm/test/CodeGen/Generic/half.ll
@@ -39,7 +39,7 @@
 ; RUN: %if spirv-registered-target       %{ llc %s -o - -mtriple=spirv-unknown-unknown           | FileCheck %s --check-prefixes=NOCRASH   %}
 ; RUN: %if systemz-registered-target     %{ llc %s -o - -mtriple=s390x-unknown-linux-gnu         | FileCheck %s --check-prefixes=ALL,CHECK %}
 ; RUN: %if ve-registered-target          %{ llc %s -o - -mtriple=ve-unknown-unknown              | FileCheck %s --check-prefixes=ALL,BAD   %}
-; RUN: %if webassembly-registered-target %{ llc %s -o - -mtriple=wasm32-unknown-unknown          | FileCheck %s --check-prefixes=ALL,BAD   %}
+; RUN: %if webassembly-registered-target %{ llc %s -o - -mtriple=wasm32-unknown-unknown          | FileCheck %s --check-prefixes=ALL,CHECK %}
 ; RUN: %if x86-registered-target         %{ llc %s -o - -mtriple=i686-unknown-linux-gnu          | FileCheck %s --check-prefixes=ALL,CHECK %}
 ; RUN: %if x86-registered-target         %{ llc %s -o - -mtriple=x86_64-pc-windows-msvc          | FileCheck %s --check-prefixes=ALL,CHECK %}
 ; RUN: %if x86-registered-target         %{ llc %s -o - -mtriple=x86_64-unknown-linux-gnu        | FileCheck %s --check-prefixes=ALL,CHECK %}
diff --git a/llvm/test/CodeGen/WebAssembly/f16.ll b/llvm/test/CodeGen/WebAssembly/f16.ll
index 0486975f6cba7..3c31d55abfadc 100644
--- a/llvm/test/CodeGen/WebAssembly/f16.ll
+++ b/llvm/test/CodeGen/WebAssembly/f16.ll
@@ -10,12 +10,11 @@ target triple = "wasm32-unknown-unknown"
 
 define void @store(half %x, ptr %p) nounwind {
 ; ALL-LABEL: store:
-; ALL:         .functype store (f32, i32) -> ()
+; ALL:         .functype store (i32, i32) -> ()
 ; ALL-NEXT:  # %bb.0:
-; ALL-NEXT:    local.get $push2=, 1
-; ALL-NEXT:    local.get $push1=, 0
-; ALL-NEXT:    call $push0=, __truncsfhf2, $pop1
-; ALL-NEXT:    i32.store16 0($pop2), $pop0
+; ALL-NEXT:    local.get $push1=, 1
+; ALL-NEXT:    local.get $push0=, 0
+; ALL-NEXT:    i32.store16 0($pop1), $pop0
 ; ALL-NEXT:    return
   store half %x, ptr %p
   ret void
@@ -23,12 +22,11 @@ define void @store(half %x, ptr %p) nounwind {
 
 define half @return(ptr %p) nounwind {
 ; ALL-LABEL: return:
-; ALL:         .functype return (i32) -> (f32)
+; ALL:         .functype return (i32) -> (i32)
 ; ALL-NEXT:  # %bb.0:
-; ALL-NEXT:    local.get $push2=, 0
-; ALL-NEXT:    i32.load16_u $push0=, 0($pop2)
-; ALL-NEXT:    call $push1=, __extendhfsf2, $pop0
-; ALL-NEXT:    return $pop1
+; ALL-NEXT:    local.get $push1=, 0
+; ALL-NEXT:    i32.load16_u $push0=, 0($pop1)
+; ALL-NEXT:    return $pop0
   %r = load half, ptr %p
   ret half %r
 }
@@ -80,50 +78,28 @@ define dso_local float @loadf(ptr nocapture readonly %a) local_unnamed_addr noun
 }
 
 define dso_local void @stored(ptr nocapture %a, double %b) local_unnamed_addr nounwind {
-; DEFISEL-LABEL: stored:
-; DEFISEL:         .functype stored (i32, f64) -> ()
-; DEFISEL-NEXT:  # %bb.0:
-; DEFISEL-NEXT:    local.get $push2=, 0
-; DEFISEL-NEXT:    local.get $push1=, 1
-; DEFISEL-NEXT:    call $push0=, __truncdfhf2, $pop1
-; DEFISEL-NEXT:    i32.store16 0($pop2), $pop0
-; DEFISEL-NEXT:    return
-;
-; FASTISEL-LABEL: stored:
-; FASTISEL:         .functype stored (i32, f64) -> ()
-; FASTISEL-NEXT:  # %bb.0:
-; FASTISEL-NEXT:    local.get $push4=, 0
-; FASTISEL-NEXT:    local.get $push3=, 1
-; FASTISEL-NEXT:    call $push2=, __truncdfhf2, $pop3
-; FASTISEL-NEXT:    i32.const $push1=, 65535
-; FASTISEL-NEXT:    i32.and $push0=, $pop2, $pop1
-; FASTISEL-NEXT:    i32.store16 0($pop4), $pop0
-; FASTISEL-NEXT:    return
+; ALL-LABEL: stored:
+; ALL:         .functype stored (i32, f64) -> ()
+; ALL-NEXT:  # %bb.0:
+; ALL-NEXT:    local.get $push2=, 0
+; ALL-NEXT:    local.get $push1=, 1
+; ALL-NEXT:    call $push0=, __truncdfhf2, $pop1
+; ALL-NEXT:    i32.store16 0($pop2), $pop0
+; ALL-NEXT:    return
   %x = tail call i16 @llvm.convert.to.fp16.f64(double %b)
   store i16 %x, ptr %a, align 2
   ret void
 }
 
 define dso_local void @storef(ptr nocapture %a, float %b) local_unnamed_addr nounwind {
-; DEFISEL-LABEL: storef:
-; DEFISEL:         .functype storef (i32, f32) -> ()
-; DEFISEL-NEXT:  # %bb.0:
-; DEFISEL-NEXT:    local.get $push2=, 0
-; DEFISEL-NEXT:    local.get $push1=, 1
-; DEFISEL-NEXT:    call $push0=, __truncsfhf2, $pop1
-; DEFISEL-NEXT:    i32.store16 0($pop2), $pop0
-; DEFISEL-NEXT:    return
-;
-; FASTISEL-LABEL: storef:
-; FASTISEL:         .functype storef (i32, f32) -> ()
-; FASTISEL-NEXT:  # %bb.0:
-; FASTISEL-NEXT:    local.get $push4=, 0
-; FASTISEL-NEXT:    local.get $push3=, 1
-; FASTISEL-NEXT:    call $push2=, __truncsfhf2, $pop3
-; FASTISEL-NEXT:    i32.const $push1=, 65535
-; FASTISEL-NEXT:    i32.and $push0=, $pop2, $pop1
-; FASTISEL-NEXT:    i32.store16 0($pop4), $pop0
-; FASTISEL-NEXT:    return
+; ALL-LABEL: storef:
+; ALL:         .functype storef (i32, f32) -> ()
+; ALL-NEXT:  # %bb.0:
+; ALL-NEXT:    local.get $push2=, 0
+; ALL-NEXT:    local.get $push1=, 1
+; ALL-NEXT:    call $push0=, __truncsfhf2, $pop1
+; ALL-NEXT:    i32.store16 0($pop2), $pop0
+; ALL-NEXT:    return
   %x = tail call i16 @llvm.convert.to.fp16.f32(float %b)
   store i16 %x, ptr %a, align 2
   ret void
@@ -170,33 +146,20 @@ define void @test_bitcast_to_half(ptr %addr, i16 %in) nounwind {
 
 define half @from_bits(i16 %x) nounwind {
 ; ALL-LABEL: from_bits:
-; ALL:         .functype from_bits (i32) -> (f32)
+; ALL:         .functype from_bits (i32) -> (i32)
 ; ALL-NEXT:  # %bb.0:
-; ALL-NEXT:    local.get $push1=, 0
-; ALL-NEXT:    call $push0=, __extendhfsf2, $pop1
+; ALL-NEXT:    local.get $push0=, 0
 ; ALL-NEXT:    return $pop0
   %res = bitcast i16 %x to half
   ret half %res
 }
 
 define i16 @to_bits(half %x) nounwind {
-; DEFISEL-LABEL: to_bits:
-; DEFISEL:         .functype to_bits (f32) -> (i32)
-; DEFISEL-NEXT:  # %bb.0:
-; DEFISEL-NEXT:    local.get $push3=, 0
-; DEFISEL-NEXT:    call $push1=, __truncsfhf2, $pop3
-; DEFISEL-NEXT:    i32.const $push0=, 65535
-; DEFISEL-NEXT:    i32.and $push2=, $pop1, $pop0
-; DEFISEL-NEXT:    return $pop2
-;
-; FASTISEL-LABEL: to_bits:
-; FASTISEL:         .functype to_bits (f32) -> (i32)
-; FASTISEL-NEXT:  # %bb.0:
-; FASTISEL-NEXT:    local.get $push3=, 0
-; FASTISEL-NEXT:    call $push2=, __truncsfhf2, $pop3
-; FASTISEL-NEXT:    i32.const $push1=, 65535
-; FASTISEL-NEXT:    i32.and $push0=, $pop2, $pop1
-; FASTISEL-NEXT:    return $pop0
+; ALL-LABEL: to_bits:
+; ALL:         .functype to_bits (i32) -> (i32)
+; ALL-NEXT:  # %bb.0:
+; ALL-NEXT:    local.get $push0=, 0
+; ALL-NEXT:    return $pop0
     %res = bitcast half %x to i16
     ret i16 %res
 }
@@ -559,27 +522,35 @@ define float @test_sitofp_fadd_i32(i32 %a, ptr %b) nounwind {
 ; DEFISEL-LABEL: test_sitofp_fadd_i32:
 ; DEFISEL:         .functype test_sitofp_fadd_i32 (i32, i32) -> (f32)
 ; DEFISEL-NEXT:  # %bb.0:
-; DEFISEL-NEXT:    local.get $push6=, 1
-; DEFISEL-NEXT:    i32.load16_u $push1=, 0($pop6)
+; DEFISEL-NEXT:    local.get $push8=, 1
+; DEFISEL-NEXT:    i32.load16_u $push7=, 0($pop8)
+; DEFISEL-NEXT:    local.set 1, $pop7
+; DEFISEL-NEXT:    local.get $push9=, 0
+; DEFISEL-NEXT:    f32.convert_i32_s $push0=, $pop9
+; DEFISEL-NEXT:    call $push1=, __truncsfhf2, $pop0
 ; DEFISEL-NEXT:    call $push2=, __extendhfsf2, $pop1
-; DEFISEL-NEXT:    local.get $push7=, 0
-; DEFISEL-NEXT:    f32.convert_i32_s $push0=, $pop7
-; DEFISEL-NEXT:    call $push3=, __truncsfhf2, $pop0
-; DEFISEL-NEXT:    call $push4=, __extendhfsf2, $pop3
-; DEFISEL-NEXT:    f32.add $push5=, $pop2, $pop4
-; DEFISEL-NEXT:    return $pop5
+; DEFISEL-NEXT:    local.get $push10=, 1
+; DEFISEL-NEXT:    call $push3=, __extendhfsf2, $pop10
+; DEFISEL-NEXT:    f32.add $push4=, $pop2, $pop3
+; DEFISEL-NEXT:    call $push5=, __truncsfhf2, $pop4
+; DEFISEL-NEXT:    call $push6=, __extendhfsf2, $pop5
+; DEFISEL-NEXT:    return $pop6
 ;
 ; FASTISEL-LABEL: test_sitofp_fadd_i32:
 ; FASTISEL:         .functype test_sitofp_fadd_i32 (i32, i32) -> (f32)
 ; FASTISEL-NEXT:  # %bb.0:
-; FASTISEL-NEXT:    local.get $push6=, 1
-; FASTISEL-NEXT:    i32.load16_u $push2=, 0($pop6)
+; FASTISEL-NEXT:    local.get $push8=, 1
+; FASTISEL-NEXT:    i32.load16_u $push7=, 0($pop8)
+; FASTISEL-NEXT:    local.set 1, $pop7
+; FASTISEL-NEXT:    local.get $push9=, 0
+; FASTISEL-NEXT:    f32.convert_i32_s $push1=, $pop9
+; FASTISEL-NEXT:    call $push2=, __truncsfhf2, $pop1
 ; FASTISEL-NEXT:    call $push3=, __extendhfsf2, $pop2
-; FASTISEL-NEXT:    local.get $push7=, 0
-; FASTISEL-NEXT:    f32.convert_i32_s $push1=, $pop7
-; FASTISEL-NEXT:    call $push4=, __truncsfhf2, $pop1
-; FASTISEL-NEXT:    call $push5=, __extendhfsf2, $pop4
-; FASTISEL-NEXT:    f32.add $push0=, $pop3, $pop5
+; FASTISEL-NEXT:    local.get $push10=, 1
+; FASTISEL-NEXT:    call $push4=, __extendhfsf2, $pop10
+; FASTISEL-NEXT:    f32.add $push5=, $pop3, $pop4
+; FASTISEL-NEXT:    call $push6=, __truncsfhf2, $pop5
+; FASTISEL-NEXT:    call $push0=, __extendhfsf2, $pop6
 ; FASTISEL-NEXT:    return $pop0
   %tmp0 = load half, ptr %b
   %tmp1 = sitofp i32 %a to half
@@ -590,17 +561,20 @@ define float @test_sitofp_fadd_i32(i32 %a, ptr %b) nounwind {
 
 define half @chained_fp_ops(half %x) {
 ; ALL-LABEL: chained_fp_ops:
-; ALL:         .functype chained_fp_ops (f32) -> (f32)
+; ALL:         .functype chained_fp_ops (i32) -> (i32)
+; ALL-NEXT:    .local f32
 ; ALL-NEXT:  # %bb.0: # %start
-; ALL-NEXT:    local.get $push6=, 0
-; ALL-NEXT:    call $push0=, __truncsfhf2, $pop6
-; ALL-NEXT:    call $push5=, __extendhfsf2, $pop0
-; ALL-NEXT:    local.tee $push4=, 0, $pop5
-; ALL-NEXT:    local.get $push7=, 0
-; ALL-NEXT:    f32.add $push1=, $pop4, $pop7
-; ALL-NEXT:    f32.const $push2=, 0x1p-1
-; ALL-NEXT:    f32.mul $push3=, $pop1, $pop2
-; ALL-NEXT:    return $pop3
+; ALL-NEXT:    local.get $push8=, 0
+; ALL-NEXT:    call $push7=, __extendhfsf2, $pop8
+; ALL-NEXT:    local.tee $push6=, 1, $pop7
+; ALL-NEXT:    local.get $push9=, 1
+; ALL-NEXT:    f32.add $push0=, $pop6, $pop9
+; ALL-NEXT:    call $push2=, __truncsfhf2, $pop0
+; ALL-NEXT:    call $push3=, __extendhfsf2, $pop2
+; ALL-NEXT:    f32.const $push1=, 0x1p-1
+; ALL-NEXT:    f32.mul $push4=, $pop3, $pop1
+; ALL-NEXT:    call $push5=, __truncsfhf2, $pop4
+; ALL-NEXT:    return $pop5
 start:
   %y = fmul half %x, 0xH4000
   %z = fdiv half %y, 0xH4000
@@ -609,16 +583,15 @@ start:
 
 define half @test_select_cc(half) nounwind {
 ; ALL-LABEL: test_select_cc:
-; ALL:         .functype test_select_cc (f32) -> (f32)
+; ALL:         .functype test_select_cc (i32) -> (i32)
 ; ALL-NEXT:  # %bb.0:
-; ALL-NEXT:    f32.const $push4=, 0x1p0
+; ALL-NEXT:    i32.const $push4=, 15360
+; ALL-NEXT:    i32.const $push3=, 0
+; ALL-NEXT:    local.get $push6=, 0
+; ALL-NEXT:    call $push1=, __extendhfsf2, $pop6
 ; ALL-NEXT:    f32.const $push0=, 0x0p0
-; ALL-NEXT:    local.get $push7=, 0
-; ALL-NEXT:    call $push1=, __truncsfhf2, $pop7
-; ALL-NEXT:    call $push2=, __extendhfsf2, $pop1
-; ALL-NEXT:    f32.const $push6=, 0x0p0
-; ALL-NEXT:    f32.ne $push3=, $pop2, $pop6
-; ALL-NEXT:    f32.select $push5=, $pop4, $pop0, $pop3
+; ALL-NEXT:    f32.ne $push2=, $pop1, $pop0
+; ALL-NEXT:    i32.select $push5=, $pop4, $pop3, $pop2
 ; ALL-NEXT:    return $pop5
   %2 = fcmp une half %0, 0xH0000
   %3 = uitofp i1 %2 to half
@@ -627,27 +600,28 @@ define half @test_select_cc(half) nounwind {
 
 define half @fabs(half %x) nounwind {
 ; ALL-LABEL: fabs:
-; ALL:         .functype fabs (f32) -> (f32)
+; ALL:         .functype fabs (i32) -> (i32)
 ; ALL-NEXT:  # %bb.0:
-; ALL-NEXT:    local.get $push3=, 0
-; ALL-NEXT:    call $push0=, __truncsfhf2, $pop3
-; ALL-NEXT:    call $push1=, __extendhfsf2, $pop0
-; ALL-NEXT:    f32.abs $push2=, $pop1
-; ALL-NEXT:    return $pop2
+; ALL-NEXT:    local.get $push2=, 0
+; ALL-NEXT:    i32.const $push0=, 32767
+; ALL-NEXT:    i32.and $push1=, $pop2, $pop0
+; ALL-NEXT:    return $pop1
   %a = call half @llvm.fabs.f16(half %x)
   ret half %a
 }
 
 define half @fcopysign(half %x, half %y) nounwind {
 ; ALL-LABEL: fcopysign:
-; ALL:         .functype fcopysign (f32, f32) -> (f32)
+; ALL:         .functype fcopysign (i32, i32) -> (i32)
 ; ALL-NEXT:  # %bb.0:
-; ALL-NEXT:    local.get $push3=, 0
-; ALL-NEXT:    call $push0=, __truncsfhf2, $pop3
-; ALL-NEXT:    call $push1=, __extendhfsf2, $pop0
-; ALL-NEXT:    local.get $push4=, 1
-; ALL-NEXT:    f32.copysign $push2=, $pop1, $pop4
-; ALL-NEXT:    return $pop2
+; ALL-NEXT:    local.get $push5=, 0
+; ALL-NEXT:    i32.const $push2=, 32767
+; ALL-NEXT:    i32.and $push3=, $pop5, $pop2
+; ALL-NEXT:    local.get $push6=, 1
+; ALL-NEXT:    i32.const $push0=, -32768
+; ALL-NEXT:    i32.and $push1=, $pop6, $pop0
+; ALL-NEXT:    i32.or $push4=, $pop3, $pop1
+; ALL-NEXT:    return $pop4
   %a = call half @llvm.copysign.f16(half %x, half %y)
   ret half %a
 }
diff --git a/llvm/test/CodeGen/WebAssembly/fpclamptosat.ll b/llvm/test/CodeGen/WebAssembly/fpclamptosat.ll
index 137994ceac132..0fbaf815b9eba 100644
--- a/llvm/test/CodeGen/WebAssembly/fpclamptosat.ll
+++ b/llvm/test/CodeGen/WebAssembly/fpclamptosat.ll
@@ -103,10 +103,9 @@ entry:
 
 define i32 @stest_f16i32(half %x) {
 ; CHECK-LABEL: stest_f16i32:
-; CHECK:         .functype stest_f16i32 (f32) -> (i32)
+; CHECK:         .functype stest_f16i32 (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    # fallthrough-return
@@ -122,10 +121,9 @@ entry:
 
 define i32 @stest_f16i32_cse(half %x) {
 ; CHECK-LABEL: stest_f16i32_cse:
-; CHECK:         .functype stest_f16i32_cse (f32) -> (i32)
+; CHECK:         .functype stest_f16i32_cse (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i64.trunc_sat_f32_s
 ; CHECK-NEXT:    i32.wrap_i64
@@ -138,10 +136,9 @@ entry:
 
 define i32 @utesth_f16i32(half %x) {
 ; CHECK-LABEL: utesth_f16i32:
-; CHECK:         .functype utesth_f16i32 (f32) -> (i32)
+; CHECK:         .functype utesth_f16i32 (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -155,10 +152,9 @@ entry:
 
 define i32 @utesth_f16i32_cse(half %x) {
 ; CHECK-LABEL: utesth_f16i32_cse:
-; CHECK:         .functype utesth_f16i32_cse (f32) -> (i32)
+; CHECK:         .functype utesth_f16i32_cse (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i64.trunc_sat_f32_u
 ; CHECK-NEXT:    i32.wrap_i64
@@ -171,10 +167,9 @@ entry:
 
 define i32 @ustest_f16i32(half %x) {
 ; CHECK-LABEL: ustest_f16i32:
-; CHECK:         .functype ustest_f16i32 (f32) -> (i32)
+; CHECK:         .functype ustest_f16i32 (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -190,10 +185,9 @@ entry:
 
 define i32 @ustest_f16i32_cse(half %x) {
 ; CHECK-LABEL: ustest_f16i32_cse:
-; CHECK:         .functype ustest_f16i32_cse (f32) -> (i32)
+; CHECK:         .functype ustest_f16i32_cse (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -373,22 +367,20 @@ entry:
 
 define i16 @stest_f16i16(half %x) {
 ; CHECK-LABEL: stest_f16i16:
-; CHECK:         .functype stest_f16i16 (f32) -> (i32)
-; CHECK-NEXT:    .local i32
+; CHECK:         .functype stest_f16i16 (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const 32767
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const 32767
 ; CHECK-NEXT:    i32.lt_s
 ; CHECK-NEXT:    i32.select
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const -32768
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const -32768
 ; CHECK-NEXT:    i32.gt_s
 ; CHECK-NEXT:    i32.select
@@ -405,16 +397,14 @@ entry:
 
 define i16 @utesth_f16i16(half %x) {
 ; CHECK-LABEL: utesth_f16i16:
-; CHECK:         .functype utesth_f16i16 (f32) -> (i32)
-; CHECK-NEXT:    .local i32
+; CHECK:         .functype utesth_f16i16 (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const 65535
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const 65535
 ; CHECK-NEXT:    i32.lt_u
 ; CHECK-NEXT:    i32.select
@@ -429,10 +419,9 @@ entry:
 
 define i16 @utesth_f16i16_cse(half %x) {
 ; CHECK-LABEL: utesth_f16i16_cse:
-; CHECK:         .functype utesth_f16i16_cse (f32) -> (i32)
+; CHECK:         .functype utesth_f16i16_cse (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -444,22 +433,20 @@ entry:
 
 define i16 @ustest_f16i16(half %x) {
 ; CHECK-LABEL: ustest_f16i16:
-; CHECK:         .functype ustest_f16i16 (f32) -> (i32)
-; CHECK-NEXT:    .local i32
+; CHECK:         .functype ustest_f16i16 (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const 65535
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const 65535
 ; CHECK-NEXT:    i32.lt_s
 ; CHECK-NEXT:    i32.select
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const 0
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const 0
 ; CHECK-NEXT:    i32.gt_s
 ; CHECK-NEXT:    i32.select
@@ -476,10 +463,9 @@ entry:
 
 define i16 @ustest_f16i16_cse(half %x) {
 ; CHECK-LABEL: ustest_f16i16_cse:
-; CHECK:         .functype ustest_f16i16_cse (f32) -> (i32)
+; CHECK:         .functype ustest_f16i16_cse (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -863,10 +849,9 @@ entry:
 
 define i64 @stest_f16i64(half %x) {
 ; CHECK-LABEL: stest_f16i64:
-; CHECK:         .functype stest_f16i64 (f32) -> (i64)
+; CHECK:         .functype stest_f16i64 (i32) -> (i64)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i64.trunc_sat_f32_s
 ; CHECK-NEXT:    # fallthrough-return
@@ -882,7 +867,7 @@ entry:
 
 define i64 @utesth_f16i64(half %x) {
 ; CHECK-LABEL: utesth_f16i64:
-; CHECK:         .functype utesth_f16i64 (f32) -> (i64)
+; CHECK:         .functype utesth_f16i64 (i32) -> (i64)
 ; CHECK-NEXT:    .local i32, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -892,7 +877,6 @@ define i64 @utesth_f16i64(half %x) {
 ; CHECK-NEXT:    global.set __stack_pointer
 ; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixunssfti
 ; CHECK-NEXT:    local.get 1
@@ -921,7 +905,7 @@ entry:
 
 define i64 @utesth_f16i64_cse(half %x) {
 ; CHECK-LABEL: utesth_f16i64_cse:
-; CHECK:         .functype utesth_f16i64_cse (f32) -> (i64)
+; CHECK:         .functype utesth_f16i64_cse (i32) -> (i64)
 ; CHECK-NEXT:    .local i32, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -931,7 +915,6 @@ define i64 @utesth_f16i64_cse(half %x) {
 ; CHECK-NEXT:    global.set __stack_pointer
 ; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixunssfti
 ; CHECK-NEXT:    local.get 1
@@ -951,7 +934,7 @@ entry:
 
 define i64 @ustest_f16i64(half %x) {
 ; CHECK-LABEL: ustest_f16i64:
-; CHECK:         .functype ustest_f16i64 (f32) -> (i64)
+; CHECK:         .functype ustest_f16i64 (i32) -> (i64)
 ; CHECK-NEXT:    .local i32, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -961,7 +944,6 @@ define i64 @ustest_f16i64(half %x) {
 ; CHECK-NEXT:    global.set __stack_pointer
 ; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 1
@@ -1103,10 +1085,9 @@ entry:
 
 define i32 @stest_f16i32_mm(half %x) {
 ; CHECK-LABEL: stest_f16i32_mm:
-; CHECK:         .functype stest_f16i32_mm (f32) -> (i32)
+; CHECK:         .functype stest_f16i32_mm (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    # fallthrough-return
@@ -1120,10 +1101,9 @@ entry:
 
 define i32 @utesth_f16i32_mm(half %x) {
 ; CHECK-LABEL: utesth_f16i32_mm:
-; CHECK:         .functype utesth_f16i32_mm (f32) -> (i32)
+; CHECK:         .functype utesth_f16i32_mm (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -1136,10 +1116,9 @@ entry:
 
 define i32 @ustest_f16i32_mm(half %x) {
 ; CHECK-LABEL: ustest_f16i32_mm:
-; CHECK:         .functype ustest_f16i32_mm (f32) -> (i32)
+; CHECK:         .functype ustest_f16i32_mm (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -1153,10 +1132,9 @@ entry:
 
 define i32 @ustest_f16i32_mm_cse(half %x) {
 ; CHECK-LABEL: ustest_f16i32_mm_cse:
-; CHECK:         .functype ustest_f16i32_mm_cse (f32) -> (i32)
+; CHECK:         .functype ustest_f16i32_mm_cse (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -1325,22 +1303,20 @@ entry:
 
 define i16 @stest_f16i16_mm(half %x) {
 ; CHECK-LABEL: stest_f16i16_mm:
-; CHECK:         .functype stest_f16i16_mm (f32) -> (i32)
-; CHECK-NEXT:    .local i32
+; CHECK:         .functype stest_f16i16_mm (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const 32767
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const 32767
 ; CHECK-NEXT:    i32.lt_s
 ; CHECK-NEXT:    i32.select
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const -32768
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const -32768
 ; CHECK-NEXT:    i32.gt_s
 ; CHECK-NEXT:    i32.select
@@ -1355,16 +1331,14 @@ entry:
 
 define i16 @utesth_f16i16_mm(half %x) {
 ; CHECK-LABEL: utesth_f16i16_mm:
-; CHECK:         .functype utesth_f16i16_mm (f32) -> (i32)
-; CHECK-NEXT:    .local i32
+; CHECK:         .functype utesth_f16i16_mm (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const 65535
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const 65535
 ; CHECK-NEXT:    i32.lt_u
 ; CHECK-NEXT:    i32.select
@@ -1378,22 +1352,20 @@ entry:
 
 define i16 @ustest_f16i16_mm(half %x) {
 ; CHECK-LABEL: ustest_f16i16_mm:
-; CHECK:         .functype ustest_f16i16_mm (f32) -> (i32)
-; CHECK-NEXT:    .local i32
+; CHECK:         .functype ustest_f16i16_mm (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const 65535
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const 65535
 ; CHECK-NEXT:    i32.lt_s
 ; CHECK-NEXT:    i32.select
-; CHECK-NEXT:    local.tee 1
+; CHECK-NEXT:    local.tee 0
 ; CHECK-NEXT:    i32.const 0
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 0
 ; CHECK-NEXT:    i32.const 0
 ; CHECK-NEXT:    i32.gt_s
 ; CHECK-NEXT:    i32.select
@@ -1408,10 +1380,9 @@ entry:
 
 define i16 @ustest_f16i16_mm_cse(half %x) {
 ; CHECK-LABEL: ustest_f16i16_mm_cse:
-; CHECK:         .functype ustest_f16i16_mm_cse (f32) -> (i32)
+; CHECK:         .functype ustest_f16i16_mm_cse (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -1622,10 +1593,9 @@ entry:
 
 define i64 @stest_f16i64_mm(half %x) {
 ; CHECK-LABEL: stest_f16i64_mm:
-; CHECK:         .functype stest_f16i64_mm (f32) -> (i64)
+; CHECK:         .functype stest_f16i64_mm (i32) -> (i64)
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i64.trunc_sat_f32_s
 ; CHECK-NEXT:    # fallthrough-return
@@ -1639,7 +1609,7 @@ entry:
 
 define i64 @utesth_f16i64_mm(half %x) {
 ; CHECK-LABEL: utesth_f16i64_mm:
-; CHECK:         .functype utesth_f16i64_mm (f32) -> (i64)
+; CHECK:         .functype utesth_f16i64_mm (i32) -> (i64)
 ; CHECK-NEXT:    .local i32, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -1649,7 +1619,6 @@ define i64 @utesth_f16i64_mm(half %x) {
 ; CHECK-NEXT:    global.set __stack_pointer
 ; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixunssfti
 ; CHECK-NEXT:    local.get 1
@@ -1677,7 +1646,7 @@ entry:
 
 define i64 @ustest_f16i64_mm(half %x) {
 ; CHECK-LABEL: ustest_f16i64_mm:
-; CHECK:         .functype ustest_f16i64_mm (f32) -> (i64)
+; CHECK:         .functype ustest_f16i64_mm (i32) -> (i64)
 ; CHECK-NEXT:    .local i32, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -1687,7 +1656,6 @@ define i64 @ustest_f16i64_mm(half %x) {
 ; CHECK-NEXT:    global.set __stack_pointer
 ; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 1
@@ -1726,7 +1694,7 @@ entry:
 
 define i64 @utesth_f16i64_mm_cse(half %x) {
 ; CHECK-LABEL: utesth_f16i64_mm_cse:
-; CHECK:         .functype utesth_f16i64_mm_cse (f32) -> (i64)
+; CHECK:         .functype utesth_f16i64_mm_cse (i32) -> (i64)
 ; CHECK-NEXT:    .local i32, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -1736,7 +1704,6 @@ define i64 @utesth_f16i64_mm_cse(half %x) {
 ; CHECK-NEXT:    global.set __stack_pointer
 ; CHECK-NEXT:    local.get 1
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixunssfti
 ; CHECK-NEXT:    local.get 1
diff --git a/llvm/test/CodeGen/WebAssembly/fpclamptosat_vec.ll b/llvm/test/CodeGen/WebAssembly/fpclamptosat_vec.ll
index 7190e162eb010..fabc5c174fd0b 100644
--- a/llvm/test/CodeGen/WebAssembly/fpclamptosat_vec.ll
+++ b/llvm/test/CodeGen/WebAssembly/fpclamptosat_vec.ll
@@ -174,27 +174,24 @@ entry:
 
 define <4 x i32> @stest_f16i32(<4 x half> %x) {
 ; CHECK-LABEL: stest_f16i32:
-; CHECK:         .functype stest_f16i32 (f32, f32, f32, f32) -> (v128)
+; CHECK:         .functype stest_f16i32 (i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 4
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 4
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
@@ -211,27 +208,24 @@ entry:
 
 define <4 x i32> @utesth_f16i32(<4 x half> %x) {
 ; CHECK-LABEL: utesth_f16i32:
-; CHECK:         .functype utesth_f16i32 (f32, f32, f32, f32) -> (v128)
+; CHECK:         .functype utesth_f16i32 (i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 4
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 4
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 3
@@ -246,27 +240,24 @@ entry:
 
 define <4 x i32> @ustest_f16i32(<4 x half> %x) {
 ; CHECK-LABEL: ustest_f16i32:
-; CHECK:         .functype ustest_f16i32 (f32, f32, f32, f32) -> (v128)
+; CHECK:         .functype ustest_f16i32 (i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 4
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 4
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 3
@@ -433,73 +424,65 @@ entry:
 
 define <8 x i16> @stest_f16i16(<8 x half> %x) {
 ; CHECK-LABEL: stest_f16i16:
-; CHECK:         .functype stest_f16i16 (f32, f32, f32, f32, f32, f32, f32, f32) -> (v128)
-; CHECK-NEXT:    .local v128, v128, v128
+; CHECK:         .functype stest_f16i16 (i32, i32, i32, i32, i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32, f32, f32, f32, f32, v128, v128, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 5
+; CHECK-NEXT:    local.set 8
 ; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 4
+; CHECK-NEXT:    local.set 9
 ; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 6
+; CHECK-NEXT:    local.set 10
 ; CHECK-NEXT:    local.get 7
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 7
+; CHECK-NEXT:    local.set 11
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 12
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 12
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
 ; CHECK-NEXT:    v128.const 32767, 32767, 32767, 32767
-; CHECK-NEXT:    local.tee 8
+; CHECK-NEXT:    local.tee 13
 ; CHECK-NEXT:    i32x4.min_s
 ; CHECK-NEXT:    v128.const -32768, -32768, -32768, -32768
-; CHECK-NEXT:    local.tee 9
+; CHECK-NEXT:    local.tee 14
 ; CHECK-NEXT:    i32x4.max_s
 ; CHECK-NEXT:    v128.const 65535, 65535, 65535, 65535
-; CHECK-NEXT:    local.tee 10
+; CHECK-NEXT:    local.tee 15
 ; CHECK-NEXT:    v128.and
-; CHECK-NEXT:    local.get 4
+; CHECK-NEXT:    local.get 9
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 5
+; CHECK-NEXT:    local.get 8
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
-; CHECK-NEXT:    local.get 6
+; CHECK-NEXT:    local.get 10
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
-; CHECK-NEXT:    local.get 7
+; CHECK-NEXT:    local.get 11
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
-; CHECK-NEXT:    local.get 8
+; CHECK-NEXT:    local.get 13
 ; CHECK-NEXT:    i32x4.min_s
-; CHECK-NEXT:    local.get 9
+; CHECK-NEXT:    local.get 14
 ; CHECK-NEXT:    i32x4.max_s
-; CHECK-NEXT:    local.get 10
+; CHECK-NEXT:    local.get 15
 ; CHECK-NEXT:    v128.and
 ; CHECK-NEXT:    i16x8.narrow_i32x4_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -515,63 +498,55 @@ entry:
 
 define <8 x i16> @utesth_f16i16(<8 x half> %x) {
 ; CHECK-LABEL: utesth_f16i16:
-; CHECK:         .functype utesth_f16i16 (f32, f32, f32, f32, f32, f32, f32, f32) -> (v128)
-; CHECK-NEXT:    .local v128
+; CHECK:         .functype utesth_f16i16 (i32, i32, i32, i32, i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32, f32, f32, f32, f32, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 5
+; CHECK-NEXT:    local.set 8
 ; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 4
+; CHECK-NEXT:    local.set 9
 ; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 6
+; CHECK-NEXT:    local.set 10
 ; CHECK-NEXT:    local.get 7
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 7
+; CHECK-NEXT:    local.set 11
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 12
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 12
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 3
 ; CHECK-NEXT:    v128.const 65535, 65535, 65535, 65535
-; CHECK-NEXT:    local.tee 8
+; CHECK-NEXT:    local.tee 13
 ; CHECK-NEXT:    i32x4.min_u
-; CHECK-NEXT:    local.get 4
+; CHECK-NEXT:    local.get 9
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 5
+; CHECK-NEXT:    local.get 8
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 1
-; CHECK-NEXT:    local.get 6
+; CHECK-NEXT:    local.get 10
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 2
-; CHECK-NEXT:    local.get 7
+; CHECK-NEXT:    local.get 11
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 3
-; CHECK-NEXT:    local.get 8
+; CHECK-NEXT:    local.get 13
 ; CHECK-NEXT:    i32x4.min_u
 ; CHECK-NEXT:    i16x8.narrow_i32x4_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -585,68 +560,60 @@ entry:
 
 define <8 x i16> @ustest_f16i16(<8 x half> %x) {
 ; CHECK-LABEL: ustest_f16i16:
-; CHECK:         .functype ustest_f16i16 (f32, f32, f32, f32, f32, f32, f32, f32) -> (v128)
-; CHECK-NEXT:    .local v128, v128
+; CHECK:         .functype ustest_f16i16 (i32, i32, i32, i32, i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32, f32, f32, f32, f32, v128, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 5
+; CHECK-NEXT:    local.set 8
 ; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 4
+; CHECK-NEXT:    local.set 9
 ; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 6
+; CHECK-NEXT:    local.set 10
 ; CHECK-NEXT:    local.get 7
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 7
+; CHECK-NEXT:    local.set 11
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 12
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 12
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
 ; CHECK-NEXT:    v128.const 65535, 65535, 65535, 65535
-; CHECK-NEXT:    local.tee 8
+; CHECK-NEXT:    local.tee 13
 ; CHECK-NEXT:    i32x4.min_s
 ; CHECK-NEXT:    v128.const 0, 0, 0, 0
-; CHECK-NEXT:    local.tee 9
+; CHECK-NEXT:    local.tee 14
 ; CHECK-NEXT:    i32x4.max_s
-; CHECK-NEXT:    local.get 4
+; CHECK-NEXT:    local.get 9
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 5
+; CHECK-NEXT:    local.get 8
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
-; CHECK-NEXT:    local.get 6
+; CHECK-NEXT:    local.get 10
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
-; CHECK-NEXT:    local.get 7
+; CHECK-NEXT:    local.get 11
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
-; CHECK-NEXT:    local.get 8
+; CHECK-NEXT:    local.get 13
 ; CHECK-NEXT:    i32x4.min_s
-; CHECK-NEXT:    local.get 9
+; CHECK-NEXT:    local.get 14
 ; CHECK-NEXT:    i32x4.max_s
 ; CHECK-NEXT:    i16x8.narrow_i32x4_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -1186,7 +1153,7 @@ entry:
 
 define <2 x i64> @stest_f16i64(<2 x half> %x) {
 ; CHECK-LABEL: stest_f16i64:
-; CHECK:         .functype stest_f16i64 (f32, f32) -> (v128)
+; CHECK:         .functype stest_f16i64 (i32, i32) -> (v128)
 ; CHECK-NEXT:    .local i32, i64, i64, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -1198,12 +1165,10 @@ define <2 x i64> @stest_f16i64(<2 x half> %x) {
 ; CHECK-NEXT:    i32.const 16
 ; CHECK-NEXT:    i32.add
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 2
@@ -1297,7 +1262,7 @@ entry:
 
 define <2 x i64> @utesth_f16i64(<2 x half> %x) {
 ; CHECK-LABEL: utesth_f16i64:
-; CHECK:         .functype utesth_f16i64 (f32, f32) -> (v128)
+; CHECK:         .functype utesth_f16i64 (i32, i32) -> (v128)
 ; CHECK-NEXT:    .local i32, i64, i64, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -1309,12 +1274,10 @@ define <2 x i64> @utesth_f16i64(<2 x half> %x) {
 ; CHECK-NEXT:    i32.const 16
 ; CHECK-NEXT:    i32.add
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixunssfti
 ; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixunssfti
 ; CHECK-NEXT:    local.get 2
@@ -1356,7 +1319,7 @@ entry:
 
 define <2 x i64> @ustest_f16i64(<2 x half> %x) {
 ; CHECK-LABEL: ustest_f16i64:
-; CHECK:         .functype ustest_f16i64 (f32, f32) -> (v128)
+; CHECK:         .functype ustest_f16i64 (i32, i32) -> (v128)
 ; CHECK-NEXT:    .local i32, i64, i64, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -1368,12 +1331,10 @@ define <2 x i64> @ustest_f16i64(<2 x half> %x) {
 ; CHECK-NEXT:    i32.const 16
 ; CHECK-NEXT:    i32.add
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 2
@@ -1616,27 +1577,24 @@ entry:
 
 define <4 x i32> @stest_f16i32_mm(<4 x half> %x) {
 ; CHECK-LABEL: stest_f16i32_mm:
-; CHECK:         .functype stest_f16i32_mm (f32, f32, f32, f32) -> (v128)
+; CHECK:         .functype stest_f16i32_mm (i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 4
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 4
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
@@ -1651,27 +1609,24 @@ entry:
 
 define <4 x i32> @utesth_f16i32_mm(<4 x half> %x) {
 ; CHECK-LABEL: utesth_f16i32_mm:
-; CHECK:         .functype utesth_f16i32_mm (f32, f32, f32, f32) -> (v128)
+; CHECK:         .functype utesth_f16i32_mm (i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 4
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 4
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 3
@@ -1685,27 +1640,24 @@ entry:
 
 define <4 x i32> @ustest_f16i32_mm(<4 x half> %x) {
 ; CHECK-LABEL: ustest_f16i32_mm:
-; CHECK:         .functype ustest_f16i32_mm (f32, f32, f32, f32) -> (v128)
+; CHECK:         .functype ustest_f16i32_mm (i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 4
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 4
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 3
@@ -1860,73 +1812,65 @@ entry:
 
 define <8 x i16> @stest_f16i16_mm(<8 x half> %x) {
 ; CHECK-LABEL: stest_f16i16_mm:
-; CHECK:         .functype stest_f16i16_mm (f32, f32, f32, f32, f32, f32, f32, f32) -> (v128)
-; CHECK-NEXT:    .local v128, v128, v128
+; CHECK:         .functype stest_f16i16_mm (i32, i32, i32, i32, i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32, f32, f32, f32, f32, v128, v128, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 5
+; CHECK-NEXT:    local.set 8
 ; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 4
+; CHECK-NEXT:    local.set 9
 ; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 6
+; CHECK-NEXT:    local.set 10
 ; CHECK-NEXT:    local.get 7
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 7
+; CHECK-NEXT:    local.set 11
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 12
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 12
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
 ; CHECK-NEXT:    v128.const 32767, 32767, 32767, 32767
-; CHECK-NEXT:    local.tee 8
+; CHECK-NEXT:    local.tee 13
 ; CHECK-NEXT:    i32x4.min_s
 ; CHECK-NEXT:    v128.const -32768, -32768, -32768, -32768
-; CHECK-NEXT:    local.tee 9
+; CHECK-NEXT:    local.tee 14
 ; CHECK-NEXT:    i32x4.max_s
 ; CHECK-NEXT:    v128.const 65535, 65535, 65535, 65535
-; CHECK-NEXT:    local.tee 10
+; CHECK-NEXT:    local.tee 15
 ; CHECK-NEXT:    v128.and
-; CHECK-NEXT:    local.get 4
+; CHECK-NEXT:    local.get 9
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 5
+; CHECK-NEXT:    local.get 8
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
-; CHECK-NEXT:    local.get 6
+; CHECK-NEXT:    local.get 10
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
-; CHECK-NEXT:    local.get 7
+; CHECK-NEXT:    local.get 11
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
-; CHECK-NEXT:    local.get 8
+; CHECK-NEXT:    local.get 13
 ; CHECK-NEXT:    i32x4.min_s
-; CHECK-NEXT:    local.get 9
+; CHECK-NEXT:    local.get 14
 ; CHECK-NEXT:    i32x4.max_s
-; CHECK-NEXT:    local.get 10
+; CHECK-NEXT:    local.get 15
 ; CHECK-NEXT:    v128.and
 ; CHECK-NEXT:    i16x8.narrow_i32x4_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -1940,63 +1884,55 @@ entry:
 
 define <8 x i16> @utesth_f16i16_mm(<8 x half> %x) {
 ; CHECK-LABEL: utesth_f16i16_mm:
-; CHECK:         .functype utesth_f16i16_mm (f32, f32, f32, f32, f32, f32, f32, f32) -> (v128)
-; CHECK-NEXT:    .local v128
+; CHECK:         .functype utesth_f16i16_mm (i32, i32, i32, i32, i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32, f32, f32, f32, f32, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 5
+; CHECK-NEXT:    local.set 8
 ; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 4
+; CHECK-NEXT:    local.set 9
 ; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 6
+; CHECK-NEXT:    local.set 10
 ; CHECK-NEXT:    local.get 7
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 7
+; CHECK-NEXT:    local.set 11
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 12
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 12
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 3
 ; CHECK-NEXT:    v128.const 65535, 65535, 65535, 65535
-; CHECK-NEXT:    local.tee 8
+; CHECK-NEXT:    local.tee 13
 ; CHECK-NEXT:    i32x4.min_u
-; CHECK-NEXT:    local.get 4
+; CHECK-NEXT:    local.get 9
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 5
+; CHECK-NEXT:    local.get 8
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 1
-; CHECK-NEXT:    local.get 6
+; CHECK-NEXT:    local.get 10
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 2
-; CHECK-NEXT:    local.get 7
+; CHECK-NEXT:    local.get 11
 ; CHECK-NEXT:    i32.trunc_sat_f32_u
 ; CHECK-NEXT:    i32x4.replace_lane 3
-; CHECK-NEXT:    local.get 8
+; CHECK-NEXT:    local.get 13
 ; CHECK-NEXT:    i32x4.min_u
 ; CHECK-NEXT:    i16x8.narrow_i32x4_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -2009,68 +1945,60 @@ entry:
 
 define <8 x i16> @ustest_f16i16_mm(<8 x half> %x) {
 ; CHECK-LABEL: ustest_f16i16_mm:
-; CHECK:         .functype ustest_f16i16_mm (f32, f32, f32, f32, f32, f32, f32, f32) -> (v128)
-; CHECK-NEXT:    .local v128, v128
+; CHECK:         .functype ustest_f16i16_mm (i32, i32, i32, i32, i32, i32, i32, i32) -> (v128)
+; CHECK-NEXT:    .local f32, f32, f32, f32, f32, v128, v128
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    local.get 5
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 5
+; CHECK-NEXT:    local.set 8
 ; CHECK-NEXT:    local.get 4
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 4
+; CHECK-NEXT:    local.set 9
 ; CHECK-NEXT:    local.get 6
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 6
+; CHECK-NEXT:    local.set 10
 ; CHECK-NEXT:    local.get 7
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 7
+; CHECK-NEXT:    local.set 11
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
-; CHECK-NEXT:    local.set 1
+; CHECK-NEXT:    local.set 12
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 1
+; CHECK-NEXT:    local.get 12
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
 ; CHECK-NEXT:    local.get 2
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
 ; CHECK-NEXT:    local.get 3
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
 ; CHECK-NEXT:    v128.const 65535, 65535, 65535, 65535
-; CHECK-NEXT:    local.tee 8
+; CHECK-NEXT:    local.tee 13
 ; CHECK-NEXT:    i32x4.min_s
 ; CHECK-NEXT:    v128.const 0, 0, 0, 0
-; CHECK-NEXT:    local.tee 9
+; CHECK-NEXT:    local.tee 14
 ; CHECK-NEXT:    i32x4.max_s
-; CHECK-NEXT:    local.get 4
+; CHECK-NEXT:    local.get 9
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.splat
-; CHECK-NEXT:    local.get 5
+; CHECK-NEXT:    local.get 8
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 1
-; CHECK-NEXT:    local.get 6
+; CHECK-NEXT:    local.get 10
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 2
-; CHECK-NEXT:    local.get 7
+; CHECK-NEXT:    local.get 11
 ; CHECK-NEXT:    i32.trunc_sat_f32_s
 ; CHECK-NEXT:    i32x4.replace_lane 3
-; CHECK-NEXT:    local.get 8
+; CHECK-NEXT:    local.get 13
 ; CHECK-NEXT:    i32x4.min_s
-; CHECK-NEXT:    local.get 9
+; CHECK-NEXT:    local.get 14
 ; CHECK-NEXT:    i32x4.max_s
 ; CHECK-NEXT:    i16x8.narrow_i32x4_u
 ; CHECK-NEXT:    # fallthrough-return
@@ -2566,7 +2494,7 @@ entry:
 
 define <2 x i64> @stest_f16i64_mm(<2 x half> %x) {
 ; CHECK-LABEL: stest_f16i64_mm:
-; CHECK:         .functype stest_f16i64_mm (f32, f32) -> (v128)
+; CHECK:         .functype stest_f16i64_mm (i32, i32) -> (v128)
 ; CHECK-NEXT:    .local i32, i64, i64, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -2578,12 +2506,10 @@ define <2 x i64> @stest_f16i64_mm(<2 x half> %x) {
 ; CHECK-NEXT:    i32.const 16
 ; CHECK-NEXT:    i32.add
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 2
@@ -2675,7 +2601,7 @@ entry:
 
 define <2 x i64> @utesth_f16i64_mm(<2 x half> %x) {
 ; CHECK-LABEL: utesth_f16i64_mm:
-; CHECK:         .functype utesth_f16i64_mm (f32, f32) -> (v128)
+; CHECK:         .functype utesth_f16i64_mm (i32, i32) -> (v128)
 ; CHECK-NEXT:    .local i32, i64, i64, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -2687,12 +2613,10 @@ define <2 x i64> @utesth_f16i64_mm(<2 x half> %x) {
 ; CHECK-NEXT:    i32.const 16
 ; CHECK-NEXT:    i32.add
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixunssfti
 ; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixunssfti
 ; CHECK-NEXT:    local.get 2
@@ -2733,7 +2657,7 @@ entry:
 
 define <2 x i64> @ustest_f16i64_mm(<2 x half> %x) {
 ; CHECK-LABEL: ustest_f16i64_mm:
-; CHECK:         .functype ustest_f16i64_mm (f32, f32) -> (v128)
+; CHECK:         .functype ustest_f16i64_mm (i32, i32) -> (v128)
 ; CHECK-NEXT:    .local i32, i64, i64, i64, i64
 ; CHECK-NEXT:  # %bb.0: # %entry
 ; CHECK-NEXT:    global.get __stack_pointer
@@ -2745,12 +2669,10 @@ define <2 x i64> @ustest_f16i64_mm(<2 x half> %x) {
 ; CHECK-NEXT:    i32.const 16
 ; CHECK-NEXT:    i32.add
 ; CHECK-NEXT:    local.get 1
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 2
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call __fixsfti
 ; CHECK-NEXT:    local.get 2
diff --git a/llvm/test/CodeGen/WebAssembly/llvm.sincos.ll b/llvm/test/CodeGen/WebAssembly/llvm.sincos.ll
index 3c10b09525573..0608a60b739f8 100644
--- a/llvm/test/CodeGen/WebAssembly/llvm.sincos.ll
+++ b/llvm/test/CodeGen/WebAssembly/llvm.sincos.ll
@@ -4,36 +4,36 @@
 
 define { half, half } @test_sincos_f16(half %a) #0 {
 ; WASM32-LABEL: test_sincos_f16:
-; WASM32:         .functype test_sincos_f16 (i32, f32) -> ()
+; WASM32:         .functype test_sincos_f16 (i32, i32) -> ()
+; WASM32-NEXT:    .local f32
 ; WASM32-NEXT:  # %bb.0:
 ; WASM32-NEXT:    local.get 0
 ; WASM32-NEXT:    local.get 1
-; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    call __extendhfsf2
-; WASM32-NEXT:    local.tee 1
+; WASM32-NEXT:    local.tee 2
 ; WASM32-NEXT:    call cosf
 ; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    i32.store16 2
 ; WASM32-NEXT:    local.get 0
-; WASM32-NEXT:    local.get 1
+; WASM32-NEXT:    local.get 2
 ; WASM32-NEXT:    call sinf
 ; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    i32.store16 0
 ; WASM32-NEXT:    # fallthrough-return
 ;
 ; WASM64-LABEL: test_sincos_f16:
-; WASM64:         .functype test_sincos_f16 (i64, f32) -> ()
+; WASM64:         .functype test_sincos_f16 (i64, i32) -> ()
+; WASM64-NEXT:    .local f32
 ; WASM64-NEXT:  # %bb.0:
 ; WASM64-NEXT:    local.get 0
 ; WASM64-NEXT:    local.get 1
-; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    call __extendhfsf2
-; WASM64-NEXT:    local.tee 1
+; WASM64-NEXT:    local.tee 2
 ; WASM64-NEXT:    call cosf
 ; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    i32.store16 2
 ; WASM64-NEXT:    local.get 0
-; WASM64-NEXT:    local.get 1
+; WASM64-NEXT:    local.get 2
 ; WASM64-NEXT:    call sinf
 ; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    i32.store16 0
@@ -44,12 +44,12 @@ define { half, half } @test_sincos_f16(half %a) #0 {
 
 define half @test_sincos_f16_only_use_sin(half %a) #0 {
 ; CHECK-LABEL: test_sincos_f16_only_use_sin:
-; CHECK:         .functype test_sincos_f16_only_use_sin (f32) -> (f32)
+; CHECK:         .functype test_sincos_f16_only_use_sin (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call sinf
+; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    # fallthrough-return
   %result = call { half, half } @llvm.sincos.f16(half %a)
   %result.0 = extractvalue { half, half } %result, 0
@@ -58,12 +58,12 @@ define half @test_sincos_f16_only_use_sin(half %a) #0 {
 
 define half @test_sincos_f16_only_use_cos(half %a) #0 {
 ; CHECK-LABEL: test_sincos_f16_only_use_cos:
-; CHECK:         .functype test_sincos_f16_only_use_cos (f32) -> (f32)
+; CHECK:         .functype test_sincos_f16_only_use_cos (i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    local.get 0
-; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    call __extendhfsf2
 ; CHECK-NEXT:    call cosf
+; CHECK-NEXT:    call __truncsfhf2
 ; CHECK-NEXT:    # fallthrough-return
   %result = call { half, half } @llvm.sincos.f16(half %a)
   %result.1 = extractvalue { half, half } %result, 1
@@ -72,62 +72,60 @@ define half @test_sincos_f16_only_use_cos(half %a) #0 {
 
 define { <2 x half>, <2 x half> } @test_sincos_v2f16(<2 x half> %a) #0 {
 ; WASM32-LABEL: test_sincos_v2f16:
-; WASM32:         .functype test_sincos_v2f16 (i32, f32, f32) -> ()
+; WASM32:         .functype test_sincos_v2f16 (i32, i32, i32) -> ()
+; WASM32-NEXT:    .local f32, f32
 ; WASM32-NEXT:  # %bb.0:
 ; WASM32-NEXT:    local.get 0
 ; WASM32-NEXT:    local.get 2
-; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    call __extendhfsf2
-; WASM32-NEXT:    local.tee 2
+; WASM32-NEXT:    local.tee 3
 ; WASM32-NEXT:    call cosf
 ; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    i32.store16 6
 ; WASM32-NEXT:    local.get 0
 ; WASM32-NEXT:    local.get 1
-; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    call __extendhfsf2
-; WASM32-NEXT:    local.tee 1
+; WASM32-NEXT:    local.tee 4
 ; WASM32-NEXT:    call cosf
 ; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    i32.store16 4
 ; WASM32-NEXT:    local.get 0
-; WASM32-NEXT:    local.get 2
+; WASM32-NEXT:    local.get 3
 ; WASM32-NEXT:    call sinf
 ; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    i32.store16 2
 ; WASM32-NEXT:    local.get 0
-; WASM32-NEXT:    local.get 1
+; WASM32-NEXT:    local.get 4
 ; WASM32-NEXT:    call sinf
 ; WASM32-NEXT:    call __truncsfhf2
 ; WASM32-NEXT:    i32.store16 0
 ; WASM32-NEXT:    # fallthrough-return
 ;
 ; WASM64-LABEL: test_sincos_v2f16:
-; WASM64:         .functype test_sincos_v2f16 (i64, f32, f32) -> ()
+; WASM64:         .functype test_sincos_v2f16 (i64, i32, i32) -> ()
+; WASM64-NEXT:    .local f32, f32
 ; WASM64-NEXT:  # %bb.0:
 ; WASM64-NEXT:    local.get 0
 ; WASM64-NEXT:    local.get 2
-; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    call __extendhfsf2
-; WASM64-NEXT:    local.tee 2
+; WASM64-NEXT:    local.tee 3
 ; WASM64-NEXT:    call cosf
 ; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    i32.store16 6
 ; WASM64-NEXT:    local.get 0
 ; WASM64-NEXT:    local.get 1
-; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    call __extendhfsf2
-; WASM64-NEXT:    local.tee 1
+; WASM64-NEXT:    local.tee 4
 ; WASM64-NEXT:    call cosf
 ; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    i32.store16 4
 ; WASM64-NEXT:    local.get 0
-; WASM64-NEXT:    local.get 2
+; WASM64-NEXT:    local.get 3
 ; WASM64-NEXT:    call sinf
 ; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    i32.store16 2
 ; WASM64-NEXT:    local.get 0
-; WASM64-NEXT:    local.get 1
+; WASM64-NEXT:    local.get 4
 ; WASM64-NEXT:    call sinf
 ; WASM64-NEXT:    call __truncsfhf2
 ; WASM64-NEXT:    i32.store16 0